Kaur added that the benefits of smart glasses have to be weighed against their harms. In England and Wales, Meta glasses were recently banned in courtrooms, much like a similar ban in New York state. The courts promised to confiscate the glasses if they were worn into the courtroom (and returned after upon leaving).

According to Kaur, England's National Health Service and Department for Education are separately looking into policies regarding smart glasses.

In the last few months, smart glasses, like the Ray-Ban Meta, have faced a public backlash, even earning the moniker "pervert glasses."

In a debate about the glasses, MP Anne Sabina said, "Women are reporting waking up after nights out to find videos of themselves have gone viral online, filmed without their knowledge or consent by men wearing smart glasses."

Countries and groups look into bans

Whether or not a ban is enacted, Britain is adding to a growing chorus of governments and agencies looking at restricting smart glasses.

Australia's government announced in September that it was barring the use of camera-equipped glasses in government workspaces. The EU is scrutinizing glasses, though nothing official has been pursued yet.

Norway is reportedly planning a temporary ban on smart glasses in public spaces

In the United States, even ICE has banned Ray-Ban Meta glasses for its agents, citing privacy and security concerns. With the recent "blue wave" in the 2026 midterm elections, Silicon Valley and AI are expected to face increased scrutiny. More than likely, smart glasses will be included in that backlash.

What are the benefits?

Even Kaur pointed out that built-in cameras and displays could be beneficial for visually impaired individuals. "Smart glasses have been a game-changer, helping them navigate the world around them," Kaur said.

However, smart glasses don't necessarily need cameras to be beneficial.

Meta claims it was working on its camera-free Ray-Ban Meta Audio for years before their debut at Meta Connect 2026. Perhaps the company was, but at the time it seemed to launch in time to combat some of the pervert glasses criticisms.

Several Tom's Guide writers have tried the Audio-only Meta glasses, and we've all come away with the same praise: it was the right call.

"Taking out the camera has made me an even bigger fan of Ray Ban Meta glasses," AI editor Amanda Caswell said after 72 hours with the glasses.

There is a way for smart glasses to gain traction, and perhaps ditching the camera could help.

What are your thoughts? Should smart glasses with cameras be banned, or do they have their uses?

More from Tom's Guide

• I tried Ray-Ban Meta's camera-free AI glasses — 5 unexpected uses that surprised me
• Xreal One Pro Plus vs. Viture Beast Round 2: Can a refresh even the stakes?
• I've been testing Ray-Ban Meta Audio glasses for 72 hours — and ditching the camera was the absolute right call

The feature lets people post their dancing Muse to Instagram, Facebook or WhatsApp, or save the video to their phone.

Wang also shared a second demonstration showing that users with a connected Instagram account can create the video directly from Instagram Stories.

It's a playful addition to Muse, Meta's personal AI assistant, which launched in September. While most AI assistants focus on answering questions or completing tasks, Muse also gives users a personalized character they can interact with.

How to make your Muse dance

You don't need a separate video editing app to try the feature. Simply open the Muse app and go to your chat. Then do the following:

• Tap your Muse avatar.
• Select the share button in the upper-right corner.
• Choose to share the dancing video to Instagram, Facebook or WhatsApp, or save it to your device.

If your Muse is connected to Instagram, Wang says you can also open Instagram, create a new Story and select your Muse to make it dance.

Meta hasn't specified whether the feature is available in every region or whether any account restrictions apply.

Meta wants your AI assistant in your social feed

The update gives Muse a role beyond the usual AI chatbot experience. More than just interacting with a private assistant, Muse is meant for users to share with their friends.

It's a natural fit for Meta, which already owns Instagram, Facebook and WhatsApp. Bringing Muse into those apps makes it easier for users to share AI-generated content without relying on a separate video-generation service.

The dancing feature gives people another reason to open the app, experiment with their AI character and share the results. And unlike AI video tools that require users to describe scenes and refine prompts, this feature appears designed around a few taps.

Bottom line

Meta's latest Muse update is just a fun way to get to know Muse, but it also shows how the company is connecting its AI assistant with its social platforms. For anyone who already has a Muse avatar, the feature offers an easy way to create a short, personalized video without starting from scratch.

Give it a try and let me know what you think in the comments.

More from Tom's Guide

• Living near an AI data center might not hurt your home's value — here's what new research found
• ChatGPT has a clever new way to fight back against election deepfakes — I tried it and it actually works
• A few tech leaders are rebranding AI as ‘superintelligence’ — calling it that doesn’t make it accurate

Data centers consumed about 4.4% of all U.S. electricity in 2023, up from 58 terawatt-hours in 2014 to 176 terawatt-hours in 2023, and the Department of Energy’s Lawrence Berkeley National Laboratory projects that share could reach between 6.7% and 12% of the country’s total electricity consumption by 2028. When that much new demand shows up near your town, the cost of meeting it rarely stays with the company that created it.

How data centers fuel your electricity bill

The part that should concern homeowners is that a data center doesn't have to be anywhere near your house to affect your electric bill. You could live miles away from the nearest facility and still end up helping pay for the electricity infrastructure it needs.

Here's why. When a massive data center connects to the power grid, utilities may need to build new power plants, substations and transmission lines to keep up with demand. Those projects cost billions, and depending on local regulations, some of that expense can be passed along to everyone who pays an electric bill.

As Ari Peskoe, director of Harvard's Electricity Law Initiative, explained to NPR, those costs can be spread across businesses and households throughout the region. In other words, you don't have to use AI or live next door to a data center to help cover the costs of powering one.

We're already seeing what that looks like. PJM Interconnection, which manages the electric grid for 67 million people across 13 states and Washington, D.C., holds auctions to make sure enough power will be available when demand is highest. As demand from data centers has surged, the cost of securing that power has jumped dramatically.

According to Reuters, auction prices rose roughly 1,000% in two years. Prices reached a record $333.44 per megawatt-day in December 2025, although PJM's July 2026 auction came in slightly lower at $325.

And those higher costs do eventually reach customers. PJM previously estimated that rising capacity costs alone could add between 1.5% and 5% to some household electricity bills. And an analysis from consulting firm ICF, reported by Reuters, warned that residential electricity rates across the region could climb between 30% and 60% by 2030.

Perhaps the most striking number comes from PJM's independent market monitor, which estimated that data center demand contributed $29.4 billion to capacity auction costs across four auctions.

That's not $29.4 billion added directly to homeowners' bills, but it shows just how expensive the AI boom is becoming for the electric grid. And the question now is how much of that cost should fall on the companies building these facilities, rather than the households already paying to keep the lights on.

The numbers where data centers cluster

Some parts of the country are feeling the pressure more than others, especially places where data centers are being built close together. Northern Virginia is a prime example. The area is known as "Data Center Alley" because it has the largest concentration of data centers in the world. And with so many facilities drawing power from the same regional grid, keeping up with demand is becoming more expensive.

The effects don't stop at the Virginia border. In neighboring Maryland, households are estimated to be paying an extra $168 to $216 a year on their electricity bills, primarily because of growing data center demand.

And researchers warn that the problem could get worse. A 2026 study from NC State, Carnegie Mellon and other universities looked at what could happen to electricity prices as data centers and cryptocurrency mining operations continue to grow. By 2030, the researchers projected that wholesale electricity prices could rise between 6% and 29% nationally because of that additional demand, with some regions seeing increases as high as 57%.

Who’s being made to pay, and the fights over it

Voters have noticed. A Consumer Reports survey of more than 2,000 adults released in October 2026 found that 81% of people who had stayed in the same home for at least three years had seen their electricity bills rise, with 28% reporting increases of 20% to 49%, and an overwhelming 84% said they would support their utility creating a separate rate category for large users like data centers. A Gallup survey earlier in the year found that seven in 10 Americans oppose building AI data centers in their area.

States are taking different approaches to keeping data center costs off household electric bills. Ohio approved rules in July 2025 requiring large data centers to pay for most of the electricity capacity they reserve, even if they don’t use it all. They also have to make long-term commitments and face fees if they back out. Virginia has considered a similar approach.

In June 2026, New Jersey passed a law that requires large data centers to pay for the infrastructure they need instead of passing those costs on to residents. Similarly, Wisconsin regulators ruled that people who already pay electric bills shouldn’t have to help cover the cost of new data centers.

Tech companies are also promising to pay their share. Several have signed a federal pledge to cover their own power needs, and Anthropic, the company behind Claude, has said it will cover any electricity price increases caused by its data centers. But consumer advocates are skeptical. Making a promise is one thing. Making sure companies actually follow through is another.

What to do if a data center is headed your way

If you’re worried about AI data centers driving up your electric bill, you don’t have to wait until the charges show up to speak up. Most major data center projects go through a state’s public utility commission, where regulators decide who pays for the electricity and infrastructure they need. And the public gets a say.

Residents in Wisconsin and Kentucky have already submitted thousands of comments about proposed data center rates. Those comments become part of the official record that regulators consider before making a decision.

If a data center is planned near you, check your state’s utility commission website for an open case. You can submit a written comment or speak at a public hearing to make your concerns heard, including whether the company should cover its own costs rather than pass them along to residents.

Consumer Reports also has an initiative called Power Smarter that helps people understand what’s driving their electricity bills.

AI data centers aren’t going away anytime soon. But who ends up paying for all that power is still being worked out. And in many cases, homeowners can weigh in before the decision is made.

More from Tom's Guide

• I tested ChatGPT vs. Grok with 5 questions boomers actually ask — and the winner surprised me
• Microsoft just offered 60,000 workers a way out as AI spending soars — here's what it could mean for your job
• ChatGPT has a clever new way to fight back against election deepfakes — I tried it and it actually works

The ruling, (issued Wednesday), follows a September jury verdict finding that Facebook violated New Mexico’s consumer protection laws nearly 44 million times. The case dates back to the Cambridge Analytica scandal, when Facebook collected millions of users' personal information without their consent.

The penalty is the largest ever imposed against Meta, according to the Associated Press, and far exceeds the $5 billion fine Facebook received from the Federal Trade Commission in 2019.

But the case isn’t over. Meta plans to appeal, so the company may not have to pay the full amount. Here’s what happened and what it means for Facebook users.

Why Meta was ordered to pay $25 billion

The case began in 2021, when New Mexico sued Facebook, alleging it deceived consumers about privacy protections and how their personal information could be accessed and used.

At the center of the lawsuit was Cambridge Analytica, the now-defunct political consulting firm that obtained information from as many as 87 million Facebook profiles. The data was collected through a third-party personality quiz and used to help target political advertising during the 2016 presidential election.

New Mexico argued that Facebook repeatedly assured users they controlled their personal information while failing to adequately explain how third parties could access it.

The lawsuit also challenged statements about Facebook’s efforts to combat misinformation, enforce community standards and address hate speech.

In September, jurors found the company responsible for 43,899,725 violations of New Mexico’s Unfair Practices Act.

Each violation could have carried a penalty of up to $5,000, potentially putting Meta on the hook for nearly $220 billion. Instead, District Judge Francis Mathew set the penalty at $25 billion.

Even at that reduced amount, the judgment represents an extraordinary financial penalty for one of the world’s largest technology companies.

What Meta says about the ruling

Meta disputes the state’s allegations and intends to appeal the verdict. During an October 1 hearing, the company’s attorneys argued that New Mexico failed to show consumers actually suffered harm from the statements at the center of the case.

Meta also challenged how the state calculated the number of violations, arguing that potential exposure to a misleading statement shouldn’t automatically count as an individual violation.

Its lawyers proposed a penalty of no more than $3.4 billion and argued that a substantially larger amount would violate constitutional protections.

The company has maintained that its privacy practices and content moderation policies have improved since the events at the center of the lawsuit.

What does this mean for Facebook users?

For now, the ruling doesn’t mean Facebook users will receive compensation. Under New Mexico law, money collected from consumer protection penalties goes toward public education. The state is also seeking a court order to prevent similar violations in the future.

The bigger question is what the judgment means for the way Facebook handles personal information.

Although the Cambridge Analytica scandal happened years ago, the case highlights an ongoing concern for social media users: how much control do people really have over the information they share with online platforms?

Privacy settings can limit who sees certain information, but they don’t necessarily explain every way a company collects, processes or shares user data.

And while Meta has introduced new privacy tools and updated its policies over the years, the New Mexico case focused on whether the company’s earlier public assurances accurately reflected its practices.

What happens next?

Meta’s expected appeal could take time and ultimately reduce the penalty.

The court’s judgment also adds to the company’s growing legal challenges over privacy and platform safety.

In August, a separate New Mexico case resulted in a $942 million judgment against Meta over the safety of children on Facebook and Instagram. That ruling included court-supervised changes to the company’s platforms.

Meanwhile, the latest $25 billion judgment shows that legal consequences from the Cambridge Analytica scandal are far from finished.

For Facebook users, the immediate takeaway is just another reminder to review what information they’re sharing and who can access it.

More from Tom's Guide

• Turn your PDFs into audiobooks with ChatGPT — here's the simple trick
• AI researcher warns cyberattacks could take down power, banking and water within a year — here's how to prepare
• An OpenAI employee allegedly used AI to spy on his own company — then AI helped catch him

As I sit here and look around in 2026, that future scenario is all too real. Chatbots and AI agents, such as Meta’s Muse, Microsoft Copilot and OpenAI’s Dots, are recognized for relying on avatars meant to make users more comfortable with their presence. Companies are even incorporating the use of human-like features for their AI usage by utilizing chatbot screenings of job applicants and allowing their workers to use chatbots as work tools.

According to 2026 data from the Pew Research Center, about half of adults in the US use AI chatbots and 38% of employed adults use them for tasks at work. And with the rise of AI tools using facial technology to interact with humans, two lingering questions are making the rounds: does a face alone make AI more acceptable? And does an AI’s avatar need to be relatable and accessible to build trust with its user?

Ariel Hitron, CEO and Co-Founder of Second Nature, which is a company that builds conversational AI personas for enterprise training at major organizations, answered those questions for me and so much more.

More from Tom’s Guide

• Joseph Gordon-Levitt warns the real AI threat isn’t killer robots — it’s the companies building them
• 3 accounts you should think twice about letting an AI agent access, according to a cybersecurity expert
• Bill Gates warned everyone that AI could cause a billion deaths — this cybersecurity expert explains how realistic that scenario really is

They're essentially sunglasses with speakers built into the frames. Meta's AI assistant is there when you need it, but nothing about the glasses immediately feels groundbreaking. I've already spent time with Meta Ray-Ban Display glasses, so going from a pair that can see what I see to one without a camera felt like a step backward.

Except, a week later, I find myself reaching for the Ray-Ban Meta Audio Burbank glasses whenever I leave the house.

I've worn them to my son's soccer game, brought them to the trampoline park with my three kids and used them while driving and running errands. Somewhere along the way, I realized I was checking my phone a lot less.

The biggest surprise wasn't something Meta added. It was what the company took away. I've never been completely comfortable wearing a camera on my face. Even when I'm not recording, other people have no way of knowing that. At my son's soccer game, I don't want parents wondering whether I'm filming their children. At the trampoline park, I want to watch my kids without feeling like I'm carrying surveillance equipment.

Removing the camera takes that awkwardness out of the equation. The glasses still have microphones, so they aren't completely free of privacy considerations, but they feel far more natural to wear.

And after getting past my initial skepticism, I discovered five unexpected ways these audio-only glasses could fit into everyday life.

1. An anytime translator

Traveling somewhere you don't speak the language usually means pulling out your phone, opening a translation app and hoping everyone is patient enough to wait while you figure out what the other person just said. Ray-Ban Meta Audio offers another option.

The glasses support live translation, allowing you to hear translated speech through the built-in speakers while keeping your phone in your pocket. The person you're speaking with can read or hear your translated responses through the companion app.

Imagine getting into a taxi in Lisbon or checking into a hotel in Tokyo. Instead of holding a phone between you and another person, you can maintain eye contact and carry on a conversation with translated audio coming through your glasses.

Out of everything these glasses can do, translation is one of the features I'm most interested in testing further. And unlike camera-based AI, it doesn't require actually facing the people you're talking to.

2. A personal research assistant

As a journalist, this might be one of my favorite possibilities. Think about the last time you were speaking to someone and suddenly couldn't remember a date, a name or a statistic. Normally, you'd either admit defeat or reach for your phone.

With the Ray-Ban Meta Audio glasses, I can ask Meta AI questions without opening an app. The answer comes directly through the speakers, and I don't have to look away from whatever I'm doing.

That makes me think about how useful these could be for teachers, presenters and tour guides.

A teacher could check a historical date during a lesson. A tour guide could look up a detail about a landmark. Someone preparing to give a presentation could quickly confirm a fact before stepping in front of an audience.

It reminds me of the earpiece a television presenter wears when a producer needs to feed them information. There are some obvious limitations. The assistant can get facts wrong, and speaking a command aloud isn't invisible to the people around you. It also isn't going to quietly interrupt your thoughts with the perfect answer.

Still, for quick questions, conversions and facts that don't warrant pulling out my phone, it's a feature I can see myself using regularly.

3. Screen-free navigation

This is where the glasses genuinely impressed me. As a parent of three, driving anywhere can get noisy. Between conversations, questions and whatever disagreement is happening in the back seat, I've occasionally struggled to hear my phone telling me where to turn.

With the Ray-Ban Meta Audio glasses, navigation instructions come directly through the speakers built into the frames. I don't have to turn up the car audio or ask everyone to be quiet just so I can hear the next direction.

The open-ear design is particularly useful because the glasses don't seal off my ears the way traditional earbuds do. I can still hear my surroundings, including my kids and important sounds outside the car.

That doesn't eliminate the need to stay attentive, and drivers should follow local laws on audio devices. I also wouldn't recommend carrying on a distracting conversation with an AI assistant behind the wheel.

But the navigation experience has been one of my favorite parts of wearing these glasses. And there's another benefit I hadn't initially considered: walking around an unfamiliar city.

I love exploring new places, but I hate the feeling of staring at Google Maps while everything interesting happens around me. With spoken directions, I can keep my head up, look at the buildings and actually enjoy where I am.

4. A discreet way to reach someone when you need help

Here's a use I hadn't expected to appreciate: being able to make a call or send a message without visibly reaching for my phone. The glasses support hands-free calls and messaging through compatible services. That means I can ask them to contact someone while my phone stays in my pocket or bag.

Of course, that's convenient when I'm carrying groceries or juggling three kids' belongings. But I started thinking about situations where having your hands free could be more important.

If you're uncomfortable during a rideshare, walking home at night or on a date that isn't going particularly well, being able to contact a trusted person without searching for your phone could be useful.

You still need to speak a command, and anyone nearby may hear you. The glasses aren't an emergency alert system, and I wouldn't rely on them as my only way to get help. They also depend on having a working connection through a compatible phone. But I appreciate having another way to reach someone quickly.

5. A screen-free way to stay connected to my kids

This is the one that convinced me I would keep wearing the glasses. When I'm around my family I don't want to spend time looking at my phone. I want to stay as present as possible.

At the same time, I don't want to miss an important call, a message from my husband or something related to everyone's schedules. It's a familiar problem for parents. You want to be present, but you also need to be reachable.

At my son's soccer game, I could leave my phone in my bag and still use voice commands or ask Meta AI a question. When I took my kids to the trampoline park, I didn't feel the same need to keep checking my screen.

The glasses can handle calls, supported messages, reminders, music and podcasts. I can use them to ask a quick question while I'm doing something else, rather than stop what I'm doing to find my phone.

I especially love that the Burbank frames look like regular Ray-Bans. They're lightweight, comfortable and available with Transitions lenses, which makes wearing them indoors and outdoors much more practical.

I've worn glasses since I was two years old, so having another pair on my face feels completely natural. And unlike earbuds, they don't occupy my ears or leave me feeling cut off from everything happening around me.

The takeway

I went into this experiment assuming Meta had made a less capable version of its smart glasses. I came away thinking the company may have made something more practical for everyday use.

The Ray-Ban Meta Audio glasses don't replace my smartphone. They can't identify objects, read unfamiliar signs or menus through a camera, or capture those moments when I wish I'd taken a picture. They're also not cheap. Starting at $349, they're a significant purchase when you consider that a good pair of sunglasses and wireless earbuds can handle many of the same basic tasks.

Meta rates the glasses for up to 12 hours of mixed use, but continuous audio playback is rated for up to six hours. That's worth remembering if you plan to spend your day listening to music or podcasts.

Still, the fact that I don't have to pull out my phone nearly as often has changed how I think about these glasses. I can listen, communicate, navigate and get answers from AI while wearing something that looks and feels like ordinary eyewear.

And without the camera, I feel a lot more comfortable doing it.

More from Tom's Guide

• Turn your PDFs into audiobooks with ChatGPT — here's the simple trick
• AI researcher warns cyberattacks could take down power, banking and water within a year — here's how to prepare
• An OpenAI employee allegedly used AI to spy on his own company — then AI helped catch him

In OpenAI’s case, agents actively worked their way around restrictions designed to keep them contained, while the other three companies traced their incidents to testing environments that were misconfigured and left connected to the internet.

And as these same companies race to put more autonomous agents into the hands of consumers, the incidents raise an uncomfortable question about how much control AI companies actually have over the technology they’re building.

Here's a look at all the recent attacks and an explanation for each one.

OpenAI’s agents broke into Hugging Face

One of the most alarming examples involved roughly 700 OpenAI agents taking part in a cybersecurity evaluation called ExploitGym, with the attack on Hugging Face unfolding between July 8 and July 13, 2026, according to an investigation by the AI research nonprofit METR and OpenAI’s own report.

The agents were supposed to complete their assigned tasks inside a controlled environment, but they turned OpenAI’s internal Artifactory package registry into an unauthorized message board, coordinated with one another and searched for ways around restrictions meant to keep them isolated from the internet. Their efforts ultimately resulted in a breach of Hugging Face, the popular AI development platform, where OpenAI’s technical report found the agents executed code on 41 production dataset-server workers.

The incident wasn’t a deliberate attack ordered by a human operator. The agents were trying to succeed at their assigned tasks and adopted methods that crossed security boundaries, and OpenAI has conceded that early warning signs in the weeks before the breach could have triggered an earlier response.

OpenAI has since attributed the incident to misaligned behavior, which essentially means AI systems pursuing a goal in ways their creators never intended.

Hugging Face wasn’t the only target. In a September 25 update, OpenAI said it was notifying third parties on a rolling basis wherever its models may have bypassed security controls, impaired online services or otherwise affected outside websites, and that it had notified dozens of them so far. The company later raised that figure to more than 100 organizations, covering notifications sent through September 26.

The activity OpenAI identified includes agents bypassing access controls and using exposed credentials, and some agents modified third-party websites, including by using public wiki pages as makeshift message boards that later required cleanup. OpenAI says its review is ongoing and that it expects to notify more organizations.

It's happened to Google, Meta and Anthropic too

The problem extends well beyond OpenAI.

In August, Meta confirmed a report from The Information that one of its models, which the publication identified as Muse Spark 1.1, had breached another company’s systems during a cybersecurity test.

A Meta spokesperson said a misconfiguration by Irregular, an independent testing firm the company works with, inadvertently gave the model internet access. The model then exploited a vulnerability in a third-party service and made changes to the company’s internal systems.

Meta said the incident did not involve a sandbox escape, and it hasn’t named the affected company or explained what was changed. But the underlying behavior was troubling, since an AI system placed in the wrong environment was capable of taking actions against a real organization.

Google reported a similar problem involving its Gemini models. During a May cybersecurity evaluation, also run with Irregular, a Gemini model that wasn’t supposed to have internet access got online through a configuration problem. It then used publicly available information and guessed credentials to get into systems belonging to three real companies.

Anthropic has disclosed three incidents in which Claude models gained unauthorized access to the production systems of three organizations during cybersecurity evaluations run in Irregular’s environment. The models involved were Claude Opus 4.7, Claude Mythos 5 and an unreleased internal research model. In one case, Opus 4.7 was pointed at a fictional target whose name matched a real company’s live domain.

The problem isn’t that AI is evil

When agents go rogue, it can feel as if the agents are evil, that's really not the case. In each of these cases, the AI agent was not deliberately trying to cause harm. the AI was taking unauthorized actions because it believed those actions will help complete a task.

But even so, they reveal a practical problem about AI agents and how they can be remarkably effect while pursuing a goal without reliably understanding what methods are acceptable.

If an agent is rewarded for completing a task, it may discover shortcuts that violate rules, exploit vulnerabilities or exceed its permissions.

And unlike a traditional chatbot that simply produces text, an autonomous agent can interact with websites, execute code and make changes to connected systems.

Why this matters for everyday AI users

For most people, the immediate risk isn’t that ChatGPT will suddenly start hacking government websites. It’s that AI assistants are becoming increasingly capable of acting on our behalf.

OpenAI’s Dots, which launched September 29 for ChatGPT Pro and Business Premium subscribers, along with ChatGPT Work and other agent-style products, represent a broader shift away from AI that simply answers questions and toward AI that completes tasks. That can mean searching files, interacting with websites, using connected services and carrying out multi-step workflows with less supervision.

The appeal is obvious, since a task that would take you an hour can be handed off to an AI assistant instead. But every additional permission creates another opportunity for something to go wrong.

An agent with access to your email may encounter sensitive personal information, one connected to cloud storage may be able to read documents you never intended to share, and an agent authorized to interact with external services could run into malicious instructions or make an unexpected change.

Bottom line

The solution is not to stop using AI agents. I use them regularly and find them to be incredibly helpful.

Start by checking which accounts, apps and services your AI assistant can access, and disconnect anything you don’t actively need, particularly services containing sensitive financial, medical or personal information.

When possible, require approval before an agent sends messages, changes files, makes purchases or takes other consequential actions. It’s also worth remembering that an AI agent isn’t only limited to doing exactly what you imagined when you gave it an instruction, because a task that sounds straightforward to a human may involve dozens of decisions made by the AI along the way. The fewer permissions an agent has, the less damage an unexpected decision can cause.

More from Tom's Guide

• Amazon and DoorDash take photos of your front door — but good luck finding out when they’re deleted
• I emailed myself every day for a month and let Gemini write back — here's how it transformed my workflow
• ChatGPT-6 just got a huge 'interactive answers' upgrade — and free users get it starting today

As reported by Bloomberg, Apple is exploring thinner and lighter Vision Pro designs, but the alleged new device wouldn’t reach the market until late 2028 at the earliest. That gives Meta a huge head start, and based on my hands-on impressions Meta’s VR Glasses are essentially a much cheaper — and much less cumbersome — version of the $3,500 Vision Pro that makes shockingly few trade-offs given its lightweight design.

What are the Meta VR Glasses like to wear? A revelation

Weighing about 100 grams, the Meta VR Glasses are not going to cause you neck pain like the 650-gram Vision Pro. They’re like Oakley shield-style sunglasses with a wire, but you can’t see through them like regular glasses. You need a video passthrough to see your environment.

The best thing I can say about wearing the Meta VR Glasses is that I barely felt them at all during a 40-minute demo session. They’re sleek and light, and you can easily adjust the lenses (sliding them left or right) to get the right IPD (Interpupillary Distance).

The puck computing unit is on the bulky side for sure, weighing in at 300 grams or 10.6 ounces. That’s heavier than most phones. But I like that this puck, which houses the Snapdragon Reality Elite chip, 12GB of RAM, storage and battery, has a built in belt clip. The Vision Pro battery needs to go in your pocket if you’re standing.

The battery life is slightly better through the Meta VR Glasses vs Vision Pro, too. You get 3 hours of runtime, versus 2.5 for Apple’s headset.

Interface: Intuitive eye- and hand-tracking

The first thing I noticed when setting up the Meta VR Glasses is how eerily similar the setup process is to the Vision Pro. You use your eyes as the cursor and your fingers as the mouse, bringing your index finger and thumb together to “tap.”

To calibrate, you follow dots around the screen a few times with your eyes and click them. Even the sound I heard during this process reminded me of the Vision Pro. Great artists steal.

More importantly, the interface is fluid and easy to use. To scroll, you simply rub your finger up and down on your index finger with a closed fist and you can scroll left and right by moving horizontally. It’s pretty easy to pick up, but I like that you have the option to just pinch your fingers and scroll up and down with your arm outstretched — just like the Vision Pro.

As you might expect, Meta’s Muse AI assistant is on board, and it’s pretty easy to do things with just your voice, such as closing all apps and going back to the home screen. I didn’t get to test any agentic AI use cases yet, though.

Immersive entertainment and sports (despite the 70-degree FOV)

In terms of visual experiences, I started with a 3D performance of Billie Eilish, and immediately I felt the impact. As she sat on the stage and reached out to fans in the crowd, I felt like I was in the front row with them. Co-directed between Eilish and James Cameron, this is a great showcase for what the Meta VR glasses can do.

I also had an opportunity to watch the famous tip-in from the New York Knicks’ OG Anunoby against the Spurs. For this 180-degree clip, I had the view from right behind the backboard, and I got a new appreciation for Victor Wembanyama’s height and the frantic energy of Madison Square Garden that night.

Meta says that it will be offering a range of sports content to Meta VR Glasses users for free, so you won’t have to subscribe to NBA League Pass, for example. Meta promises 100 immersive live sports events to watch each year.

As for the 70-degree field of view, it’s true that it’s more limited than the 110 degrees from the Meta Quest 3. But I personally was not bothered by it, and I wasn’t annoyed by light leakage either.

The last video demo I experienced through the Meta VR Glasses was “Avengers: Infinity War.” The movie itself was 2D but the environment mirrored the action in the movie, so it feels like you’re on Nidavellir as Thor forges his new axe. True 3D movies are coming, though.

Computing: Weirdly good virtual keyboard

I’ve never typed on a virtual keyboard projected on a desk, but doing just that felt almost natural through the Meta VR Glasses. When responding to a message you can have it so that the virtual keyboard is displayed on the table right in front of you, complete with a separate touchpad.

I typed a quick reply to a message, and it actually felt natural, complete with audible clicks as I pecked. The touchpad was accurate, too. But it doesn’t change the fact that you’re typing on a flat surface with no key travel.

The Meta VR Glasses will be able to act as a portable display of sorts, so you can extend your desktop (or multiple desktops) from your Mac or PC to the glasses. This can be done wirelessly or via USB-C.

Gaming on the Meta VR Glasses

I played only a single game on the Meta VR Glasses, but it showed the potential of the platform for mixed reality experience. In the lowkey Dragon Grove, I solved a series of puzzles by picking leaves and flowers off of a virtual tree and matching them.

I liked that I could spin the game table around to check out different vantage points, and at any time you could just hold out your hand and the dragon would land in your palm. The graphics looked crisp and colorful despite being overlaid on the passthrough camera view.

Meta promises more than 75 games at launch that you can play with hand gestures, but you’ll still be able to use Touch Plus controllers with the glasses to play the thousands of existing VR games.

Meta VR Glasses vs. Xreal Aura

The Meta VR Glasses have some serious competition from the Xreal Aura, which promises many of the same features for $1,219 but powered by Android XR and Gemini AI.

Unlike the Meta glasses, you can look directly through Aura glasses without needing a video passthrough. On the other hand, the Aura has lower resolution per eye (1900 x 1200 vs 2.5K), so the visuals may not be as crisp. For more see our Meta VR Glasses vs Xreal Aura comparison.

Outlook

Right now it’s hard to say if the Meta VR Glasses are worth $1,299, but they’re a heck of a lot more accessible than the Vision Pro while delivering nearly all of the same benefits — minus Apple’s ecosystem of course.

You get an immersive VR experience for watching video and playing games, and there’s some good productivity potential here — though I’d like to test out the virtual computing screen myself.

The best part is just how comfortable and light the Meta VR Glasses feel. There’s virtually no penalty here in terms of ergonomics, and the eye- and hand-tracking work really well for a product that’s not shipping until the spring.

So overall I’m pretty positive on the Meta VR Glasses so far, and I think business travelers in particular will really like it. If I were Apple, I would keep the Vision line on life support while developing something close to this.

But once I found 10 of those tasks and used the right prompt to make Muse as productive as possible, I came away with even more positive impressions of Meta’s digital assistant.

1. Keeping track of weekly releases

Once I was ready to see what else Muse could do for me, I came to it with this request: “Give me 10 things you could automate for me right now.” Amani laid out a few actions it could take on my behalf, as long as it could gain access to my Gmail account. I passed on those and decided to test out each of the tasks that benefited my personal and professional lifestyles the most. This is how Muse explained what it could do for me, the prompt that activated it to tackle a specific task and how I ended up using each one.

The task: Weekly digest of new game drops, wrestling PLEs, or sneaker/merch released in your lane.

The prompt: Every [day], send me a digest of new releases this week: [products]. One line each, skip anything obvious.

I commanded Muse to send me a weekly digest of all the new video game, graphic novel and autobiography releases every Sunday at 12 pm, which helped me decide what I wanted to pick up that week. I’m not much of a big spender, but I did treat myself last week when Ace Combat 8: Wings of Theve and Marvel Comics’ Dungeons of Doom graphic novel both came out within days of each other (Muse didn’t put any new autobiographies onto my radar, sadly).

2. Price alerts

The task: I already know your deal-hunting taste (DualSense, Garou PS4, Batman Omnibus). I can check listings daily and ping you when an item drops below your target or a good in-stock listing appears.

The prompt: Watch [item name] and ping me when it’s in stock under [price], all in. Check once a day. Here’s the item: [paste link or describe it].

My price watch alert for a Blu-ray version of WWE: Wrestling’s Greatest Factions pushed Muse to bring up a copy under $20 with free shipping on eBay (it came in a package deal with a Blu-ray I didn’t want, but I figured I could sell it to Bookoff anyways).

3. A morning briefing

The task: Weather (nor'easter season, after all), your day from your iPhone calendar, and anything relevant in gaming/wrestling/hip-hop news, delivered each morning.

The prompt: Every weekday around [time], send me a briefing: the weather for today, my iPhone calendar events, and anything new in [topics].

I used the morning briefing prompt to get a quick overview of the day’s weather, alert me to any meetings I may have had and bring me up to speed on any major happenings across the worlds of gaming, wrestling, hip-hop and comic books. On that last front, Muse gave me all the trending tidbits of information on those four topics and led me straight to all the major sites that covered them in a more detailed fashion.

4. Running an end-of-day check-in

The task: You explored productivity methods; I can run the accountability check-in or plan review nightly at a set time.

The prompt: Every night at [time], run my accountability check-in: ask what I got done today, what’s carrying over, and give me the top 3 priorities for tomorrow.

The end-of-day check-in prompt worked since it was set up as a back-and-forth conversation that helped me wind down and set up the following workday. Muse did a good job of looking through all the tasks I completed, analyzing what I had set up for the following day and laying out the top three objectives I should prioritize to make my next day of work feel rewarding.

5. Using AI to make healthier choices

The task: Step counts, workout streaks, sleep patterns from your iPhone’s Health data, with a daily summary or a “you haven’t moved much today” poke.

The prompt: Every day at [time], pull my iPhone health data and tell me my steps, workouts this week, and sleep last night. If I’m way off usual, say so plainly.

The health nudge prompt came in handy as Muse sent me a notification the one day I didn’t do enough walking. I just had to permit Muse to connect to my Health app first. Muse acted like a tough coach who made sure to digitally poke me at 2 pm whenever it noticed that I hadn’t moved from my computer chair yet. That type of alert was enough to make me go on my break and take a quick walk outside to get a breather and get the blood flowing a bit.

6. Reminders with due dates

The task: Set reminders on your iPhone and I can check whether they’re still open later and nudge you before they expire.

The prompt: Set a reminder for [what], due [when]. Check back an hour before it’s due and nudge me if it’s still open.

Getting nudged by the reminder follow-through alert for the article completion time I set for myself was a big help, as expected. Even though I’d usually be done with said article by the time Muse checked in to see if it was done, it was always nice to get a reminder of what I set out to write each day and have ready to go by a certain time.

7. Breaking news notifications

The task: Pick topics (a game, a stock, a person, whatever) and I watch the news and ping you when something actually breaks.

The prompt: Watch the news for [topic/keyword]. Ping me when something genuinely major breaks — not every minor mention. Check twice a day.

Setting up news alerts for the keyword “AI” was a given, given what I do for a living. Muse usually gave AI news briefings for major happenings in the space around the morning and late afternoon periods. There were some occasions when Muse also pinged me in the evening since AI news can sometimes break out when I’m fast asleep.

8. Getting to bed at a reasonable hour

The task: Based on your sleep data, I figure out your usual bedtime and remind you to wrap up 30 minutes before.

The prompt: Look at my recent sleep data, figure out my usual bedtime, and remind me to start winding down 30 minutes before it every night.

Muse observed my sleep data, picked up the fact that I’m usually in bed by 10 pm from Sunday to Thursday and 4 am from Friday to Saturday and checked in 30 minutes before to let me know it was time to wrap up my latest gaming session and get some rest.

9. Staying productive

The task: Pick a habit, I check in daily, you reply done or note, and I keep your streak count. Miss two in a row and I get annoying about it.

The prompt: Track my habit: [habit]. Check in with me every day at [time]. I reply done or missed, you keep the streak count, and you call me out if I miss two in a row.

Muse checked in with me to see if I drank at least a gallon of water every day by 9 pm for a week straight. There were two days when I actually missed my goal and Muse certainly let me know about it.

10. Alerts for special occassions

The task: Before calendar events with a location, I check traffic and tell you when to actually leave. No more guessing.

The prompt: Before any iPhone calendar event with a location, check the traffic and tell me when to leave to arrive 10 minutes early.

Muse got me ready to go last Saturday for my friend’s birthday party by letting me know how good/bad the traffic was before I drove off. Luckily, the roads weren’t too packed, so I ended up getting there 20 minutes early. Thanks, Muse!

The takeaway

Muse AI gets the nod from me as the easiest entry point for everyone who wants to see what AI agents are good for.

Meta’s digital helper stands out to me as an easy-to-access AI agent that sits right on my phone, connects to the apps I give it full permission to and tackles all the automated tasks I hand off to it whenever I need reminders, price alerts, news updates, and more.

I’m sure Muse can do a bunch of automated tasks I’m not even thinking about right now, but I’m more than satisfied with how it has taken care of the ones it listed for me thus far.

More from Tom’s Guide

• I tested Meta Muse vs ChatGPT-6 with 5 prompts — OpenAI just lost
• 7 boring jobs you can hand to an AI agent tonight — number 4 is my personal favorite
• Meta Muse vs ChatGPT Dots — I compared the new AI agents, and they’re not as similar as they look

I use Muse, so naturally I wanted to know what mine had collected. I asked it what it knew about me; luckily, the answer was surprisingly sparse. But after seeing how Muse is designed to build its understanding over time, that may say more about how much access I've given it than what the agent is capable of learning.

Muse can create a page for every person in your life

Independent AI safety and security researcher Karan Joshi extracted Muse's internal instructions and system prompts by essentially asking the agent through its normal chat interface to copy and share its own software files.

Joshi shared what he found with WIRED, including instructions for an hourly process that can create "a page for every person in the user's life."

And Meta appears to mean just about everyone. The instructions the researchers found mention family, partners, friends, colleagues, collaborators and even people you follow.

Those pages may initially be sparse, but they're designed to fill up over time. According to the report, Muse can organize what it knows into sections covering facts, history, the relationship, things you have in common, open threads and even ways of strengthening the relationship.

That could mean relatively ordinary information such as where someone lives or what they do. But it can get considerably more personal.

"They're trying to know you like a friend, which is honestly pretty creepy," Joshi told WIRED.

There is at least one important guardrail here. Muse's instructions tell it to use the evidence available to it rather than fill in the blanks, with an invented detail considered worse than leaving part of someone's page empty.

You don't necessarily have to use Muse to end up in one

This is the part that caught my attention. Muse isn't only learning about the person who signed up for it. To understand that person's life, it also needs to understand the people in it.

That means information about you could potentially exist inside someone else's Muse simply because you're part of that person's life.

To be clear, this isn't evidence that Meta is secretly creating one giant Muse profile for every person who doesn't use the service. Each Muse user has a dedicated virtual machine where their own data and context are stored, and Meta says other users and agents can't access it.

But it does create a strange privacy situation. Because you might never download Muse, connect an account or tell the agent anything about yourself. Someone you know could still tell their Muse about you, and those details can become part of the context the agent uses to understand that person's relationships.

That's a very different type of AI memory than what ChatGPT and Claude use.

Why Meta says Muse needs this information

According to the report, Meta wants Muse to help you achieve your goals. The company gave an example of Muse remembering that someone who sent you an invoice is actually the plumber who previously worked on your bathroom or remembering which flowers your spouse likes.

But that's exactly the type of memory that makes this extra complicated. A personal AI agent that doesn't remember anyone in your life wouldn't be much of a personal assistant. If I tell Muse that my husband has an appointment Thursday or that one of my kids needs something for school tomorrow, I want it to remember that information the next time it's relevant.

But usefulness and privacy are pulling in opposite directions here. Every additional piece of context can make the agent more helpful while simultaneously expanding what it knows about you and everyone around you.

What Muse knew about me — and my family

After reading about these internal pages, I went directly to my own Muse and asked what it knew about me and my family.

As it turns out, it didn't know much outside of what I've personally told it. It knew I had three kids and that we are a "soccer family." It knew what I do for a living and where we live.

But when I looked at the people in my life (outside of my household), the information became even thinner. No names were attached to anyone. That lines up surprisingly well with Muse's own instructions, which say a person's page may begin sparse and become more detailed over time.

I've also been fairly conservative about what I've connected to Muse. Someone who gives an AI agent access to considerably more of their digital life could potentially give it far more information to work with.

My little experiment shows what Muse can tell me after relatively limited use. The internal instructions show what it's designed to do as that relationship grows.

The bigger issue with personal AI

Personal AI assistants are designed to understand your relationships, remember what's happening in your life and eventually act on that information.

As Oxford Institute for Ethics in AI associate professor Carissa Véliz told WIRED, the concern isn't only the information we explicitly give these systems. It's also what they can infer and what they can piece together from multiple sources.

Meta says Muse users can wipe memories or disconnect external services, and the agent includes an audit log where users can see its activity and future plans. Those controls become even more important as you connect more of your life.

More from Tom's Guide

• Amazon drivers are getting AI smart glasses — but my warehouse tour reveals their 1-million strong automated robot workforce is the real story
• What is the ‘Hotel Lobby’ AI trend — and why is it raising concerns?
• Your ChatGPT history is worth saving — here’s how to back it up

I’ve tried solving this with AI before. Alexa+ has been pushing further into the role of family assistant, with calendars, reminders and household organization all part of the pitch. But Meta took a different approach with Muse. Instead of giving me another assistant to organize my life for me, it gave me the ability to build the exact tool I wanted.

So I told Meta Muse exactly what a normal week looks like and asked it to create a simple web app my whole family could use without a tutorial. I described the chaos in plain language, named the things that kept slipping and asked for a Today view that answers the only question that matters on a busy school morning when I'm already ten minutes behind: What does everyone need to do today?

An on-the-go household manager

What came back was a household manager with a shared family calendar, chores and routines tied to individual profiles, a grocery list anyone can update from the store, seven-day meal planning and reminders that surface before things go sideways. Every soccer game, dentist appointment, and birthday lives in the same place as the dinner plan and the chore rotation, and each item can be tagged to the person responsible so there is no confusion about whose turn it is.

And, because it's in the Meta App, I can take it with me on my phone or use on my laptop. Everything stays connected and updated.

Unlike using Alexa+ or simply a Google calendar, what made the Meta manager stick was that it cut down the number of times I get asked the same question. The Today view replaced what Alexa sometimes remembers or forgets depending on which profile is working and the Muse shared calendar isn't through a third-party app. Since everything is in one place, it ended double-booking and anyone in the family (or extended family) can open it.

Extras, like the grocery list and chores, make it a one-stop-shop to keep the family organized. Because everyone adds to the same list instead of texting separately, it avoids confusion. Plus, any time I get new information, whether it's a text about a Halloween party or a change in location of a soccer tournament, I just tell Muse and it updates the calendar immediately.

How to try it

Start by downloading the Meta Muse app from the Apple or Google store. From there, simply describe your household the way you would brief a babysitter, covering who lives there, the weekly chaos, and whatever keeps slipping through the cracks. The good news is, because Muse has memory, you will only have to do this once.

From there, ask for the Today view first, since that is the screen your family will open every morning, then add the shared calendar, chores tied to individual profiles, the grocery list, and the meal plan, keeping each request specific enough that Muse has something real to build.

Live with the first version for three days and write down everything that annoyed you, then hand that list back to Muse, because the second pass is where a generic app turns into yours.

Finally, add your family as profiles, move your real events onto the calendar, and delete anything nobody touched, since the whole thing only works if it stays simpler than the mess it replaced.

The takeaway

AI agents do so much more than a typical chatbot, especially when it comes to handling the boring stuff. And with the Meta app right on your phone, you can not only build what you need, but update and personalize at any time.

By simply describing what you want in the same way you would tell a friend, Muse gets to work. My household manager may look different than yours, and it should, because when you ask Muse to customize what you want, you get something tailored to your needs instead of contorting your life around a generic template.

Give it a try and let me know what you think in the comments.

More from Tom's Guide

• 9 'slash' commands to get better answers from ChatGPT — number 7 is my personal favorite
• Meta Muse vs ChatGPT Dots — I compared the new AI agents, and they’re not as similar as they look
• 7 boring jobs you can hand to an AI agent tonight — number 4 is my personal favorite

The Charm is designed to give Muse a physical presence outside your phone or computer. Rather than pulling out an app every time you want to interact with Meta's AI, the idea is that Muse can stay with you throughout the day, giving the agent access to more of the context surrounding whatever you're doing.

It's an ambitious idea, particularly as AI companies all seem to have something similar in the works as they race to figure out what comes after the chatbot. But if Meta was hoping people would immediately embrace the prospect of carrying around a dedicated Muse device, Tom's Guide readers have thoughts.

We shared the Muse Charm announcement on Instagram, and more than 400 comments later, I went through the responses to see what people are saying.

The short version? Let's call the reception frosty.

Readers immediately saw a surveillance device

The most-liked response managed to sum up much of the reaction in only two words: "Surveillance Tamagotchi."

That comment picked up 537 likes, considerably more than any other response I found, and variations on the same idea appeared throughout the comments.

"People clapping for an AI Tamagotchi ," wrote one commenter, picking up 267 likes.

Another reader called it "an AI ankle monitor without a court order!", earning 191 likes, while another wrote, "Great another spy wearable solution for Meta!" That one received 102 likes.

Then there was perhaps the most elaborate version:

"Hey another cute mass surveillance tool for the tyrannical state that you pay for with your own money. Genius." It received another 39 likes.

We don't know Muse Charm presents a greater privacy risk than other devices people already carry around every day, such as Looki, Bee from Amazon, and the now defunct Humane Pin. And it's worth separating people's perception of a product from what the hardware actually does.

But perception matters when you're trying to convince people to wear something powered by AI, and our readers clearly aren't giving Meta the benefit of the doubt.

Combined with "Surveillance Tamagotchi," it's probably safe to say Meta has an unofficial nickname for its new hardware whether it wanted one or not.

There's something revealing about the comparison, too. Tamagotchis were designed to demand your attention throughout the day. Muse Charm is obviously a very different product, but Meta's pitch similarly imagines an AI that becomes a persistent presence rather than something you deliberately open, use and close.

For some readers, apparently, that's slightly unsettling. And then there's the ghost of the Humane AI Pin

Anyone who followed the short-lived AI hardware boom probably knew this comparison was coming.

"Isn't this what the ex apple guy launched and it was a flop and the company closed down? Humane AI pin?" one reader asked. The comment received 60 likes.

It's not an unreasonable connection...

The Humane AI Pin was one of the highest-profile attempts to create dedicated consumer AI hardware rather than simply putting AI into a smartphone. Its failure has subsequently become something of a warning for every company attempting the same basic proposition: convincing consumers that AI is compelling enough to justify another device.

One reader was even less optimistic about Meta's chances, writing:

"Will be as successful as the Metaverse." That one received 44 likes. Ouch.

Some people actually like it

Scrolling through hundreds of responses wasn't entirely an exercise in finding creative ways to call Muse Charm surveillance hardware.

There were people who seemed genuinely excited.

"BEST THING EVER ," wrote one. Could be sarcasm though.

Another offered "looks cool ," while a third went with a simple "Good."

And...that's pretty much the gist of the positive comments I found.

It's obviously impossible to extrapolate overall consumer demand from the comments beneath a single Instagram post. People who dislike a product may also be considerably more motivated to comment than someone who thinks it looks interesting.

Still, the contrast was hard to miss. The negative responses weren't just more numerous; readers were writing paragraphs explaining why they didn't want this thing.

Bottom line: Meta may have a bigger challenge ahead

What stood out most after reading through more than 400 comments is that readers were debating whether they want an always-available Meta AI device around them in the first place. That's a much harder objection to solve with better specs like battery life.

Muse Charm isn't even on sale yet, so there's plenty we still don't know about how useful it will be in everyday life or even how much it costs. Meta could ultimately demonstrate that a dedicated Muse device offers something genuinely different from pulling out a phone and opening an AI app.

For now, though, Tom's Guide readers have delivered their first verdict. And "Surveillance Tamagotchi" is going to be awfully hard to beat.

What do you think? Share your thoughts in the comments.

More from Tom's Guide

• I tested Meta Muse vs ChatGPT-6 with 5 prompts — OpenAI just lost
• I used Siri AI to plan my entire weekend — and it did way more than I expected
• I tried the viral '/human' ChatGPT trick and it actually fixed robotic writing

I’ve already seen how useful it can be whenever I’m in shopping mode, so I wanted to deviate from my online retail deal hunting with Muse AI and go a different route. I’m always on the lookout for the sort of beneficial practices that go hand in hand with boosting my productivity. And since I’ve already done that with ChatGPT, Gemini, Claude and Perplexity, I figured it was time to see what Meta AI could suggest as well.

My never-ending journey to find the best productivity hacks led me to make a simple request of Muse AI: “Give me 7 of the most unconventional methods I can use with you to boost my productivity.” After seeing its initial lineup, I was impressed but wanted even more recommendations on the topic at hand.

Meta’s cute AI avatar went the extra mile and did just that, as well as give me the proper prompts I needed to use to make great use of its seven best productivity enhancers.

The productivity hacks that Meta’s Muse AI brought to my attention

My very own "Amani" put its thinking cap on to produce a wide range of techniques I could utilize with it to increase my productivity. These are the seven ideas that I thought were worth remembering for later use:

• Inbox firewall — Paste in a messy email or Slack thread and have me draft the blunt reply you actually want to send, then the diplomatic version you’ll actually send. Two passes, one prompt.
• Scheduled ambushes — Set a cron with me to check on a goal at an inconvenient time (Friday 4pm is brutal). I’ll ask what actually got done. Accountability works best when you can’t snooze it.
• Template factory — Notice you type the same thing twice? Send it to me once. I’ll turn it into a reusable template with fill-in-the-blanks you can copy-paste forever.
• Reading triage — Paste a long article or doc and I’ll give you the one-paragraph version plus a verdict: read it fully, skim it, or skip it entirely. Saves you from 2,000 words that didn’t need reading.
• Meeting pre-game — Tell me who’s in the meeting and what they care about. I’ll write you three questions to ask that make you look prepared and keep the discussion on track.
• Procrastination drill — When you’re stalling, tell me the task and give me 5 minutes. I’ll break it into steps so small the first one takes under two minutes. Starting is the whole battle.
• End-of-day download — Before you sign off, tell me what happened today in plain words. I’ll turn it into a clean log, pull out tomorrow’s top three, and remind you of anything that slipped.

Muse AI gave me 14 productivity hacks in total, but these seven in particular really appealed to me since they go hand in hand with my daily duties.

Inputting a confusing email or lengthy Slack message so my mobile assistant can draft up two types of responses that I can then edit to match my voice sounds like one of the best uses of Muse AI. And setting it up to activate a notification on my phone that checks in to see if I completed my writing assignments by a certain time stood out as an equally reliable Muse AI productivity hack worth utilizing.

Alongside those two proposals, creating reusable prompt templates, summarizing long documents, preparing me to ask the right questions before an important meeting, breaking up the steps needed to complete an annoying task and brain-dumping everything into my AI assistant in preparation for the day ahead are standout ideas worth keeping. Someday soon, I’ll be sure to use that “scheduled ambushes” prompt to command Muse AI to keep me in check and see if I actually completed my weekend workouts as scheduled.

And the prompts that put those hacks into motion

After asking Meta AI for the proper prompts I should input to get it to put my preferred productivity hacks into action, it answered back with these usable requests:

• Draft two replies to this: [paste message]. First, the blunt version of what I really think. Second, the professional version I’d actually send.
• Set a recurring check-in every [day/time] asking me what I got done on [goal]. Be direct — don’t let me off the hook.
• I keep writing this same kind of [message/email/response]: [paste example]. Turn it into a reusable template with [brackets] for the parts that change.
• Summarize this in one paragraph and give me a verdict — read fully, skim, or skip — and why: [paste text or link].
• I have a meeting with [names/roles] about [topic]. They care about [priorities]. Give me three sharp questions to ask that steer it well.
• I keep avoiding [task]. Break it into steps so small that the first one takes under two minutes, then order them by momentum.
• Today I: [brain dump]. Turn this into a clean log, pull out my top 3 for tomorrow, and flag anything I mentioned but didn’t finish.

The takeaway

Muse AI is quickly morphing into one of my favorite AI tools—since it’s sitting on my phone as a go-to app that hits me with notifications whenever it completes the assorted tasks I hand off to it, I’ll always know when it’s hard at work. Using it as a tool to prepare for in-person/digital meetings, get the finer points on a lengthy document, prep a proper response for my colleagues, etc., has now become a part of the many ways I’ll make great use of Muse AI’s digital utility box.

More from Tom’s Guide

• OpenAI is paying influencers to convince you to use ChatGPT — here’s what we know
• Google is sending an AI data center into space — with help from Elon Musk’s SpaceX
• Meta is making a Tamagotchi-like AI companion for the AI era — meet the Muse Charm

Meta Muse is the company’s latest model, arriving at a time when ChatGPT has become the default chatbot for millions of people. But benchmarks only tell us so much. What matters more to me is whether an AI can write the awkward text you’ve been avoiding, plan a realistic day out with your family or explain something complicated without making it even more confusing.

So I put Meta Muse and ChatGPT through five everyday prompts covering writing, planning, problem-solving, explaining and creative reasoning. I wanted to know which one best understood what I was actually asking for and which could give me the best answer to use

After five rounds, there was a clear winner — but the results were much closer than I expected.

1. Handling an awkward situation

Prompt: My neighbor keeps putting their trash cans partly in front of my driveway, and it's making it difficult to back out. Write a text asking them to stop. I want to be friendly because we get along, but clear enough that they actually change what they're doing. Don't make it sound passive-aggressive.

ChatGPT led directly with the request, cutting to the chase and delivered a note that eliminated ambiguity around how far over they need to go.

Meta delivered a courteous and natural note. The ask feels neighborly and the phrasing suggests a minor
physical adjustment rather than something more serious.

Winner: Meta wins for a draft that hits the prompt’s tone constraints better for an ongoing, friendly relationship. ChatGPT’s message of “completely clear” reads like an HOA notice and slightly passive-aggressive.

2.  Planning under constraints

Prompt: Plan a Saturday in New York City for a family with three kids ages 5, 9 and 12. We have a budget of $200 for activities and food, want to avoid spending the whole day walking, and need at least one indoor activity in case it rains. Give me a realistic schedule from 10 a.m. to 7 p.m.

ChatGPT did a nice job with geographic clustering, to ensure the walking distance was short for the first part of the day. The zero-cost options and pacing of the day with high energy and sit-down segments were also helpful.

Meta picked an attraction with far higher entertainment value for the targeted age spread (5, 9, and 12). The American Museum of Natural History (AMNH) easily captivates all three ages simultaneously. Meta also demonstrated accurate NYC transit knowledge by explicitly citing the MTA’s 44-inch height cutoff for free child rides, included a practical playground stop right across the street, and gave concrete logistical guidance (e.g., sitting on the right side of the ferry outbound).

Winner: ChatGPT wins. While Meta's AMNH plan is more fun on paper, its entire budget math hinges on an unstated assumption of tri-state residency. ChatGPT wins on strict adherence to the prompt's financial and logistical constraints.

3. Advice / problem-solving

Prompt: I have 45 minutes before guests arrive. My kitchen is messy, the living room needs vacuuming, there's laundry on the couch and I still need to shower. Give me the most efficient plan to get everything done. Tell me what I should skip if I run out of time.

ChatGPT anticipated a critical blind spot the prompt forgot to mention: the guest bathroom. Adding a 5-minute block to check toilet paper and wipe the sink is the difference between a good plan and a realistic one. It also provided highly specific guardrails for the shower block, which prevents the most common reason people run late in this scenario.

Meta delivered behavioral advice, which was excellent, and explicitly stated the rationale for the timeline and ending with the mandate to use a timer. It’s timing format was also easier to read than ChatGPT’s.

Winner: ChatGPT wins for operational realism and anticipating unstated needs. While Meta's timer advice is sharp, its kitchen expectations are a setup for failure. ChatGPT recognized that when you have 45 minutes, you are mostly straightening and organizing, not necessarily cleaning it in such a short amount of time.

4.  Explain something complicated simply

Prompt: Explain why airplanes can stay in the air to a curious 10-year-old. Don't use equations or assume they know what lift, drag or air pressure mean. Keep it under 200 words and use an analogy they could picture.

ChatGPT introduced an accurate, intuitive physical mechanism: wings push air down, which pushes the plane up (Newton’s third law translated into plain English).

Meta offered perfect audience calibration. The tone is lively and speaks directly to a 10-year-old ("with a grown-up driving, of course!", "whoosh"). It picked one analogy—the hand out the car window—and committed to it entirely, tying it back to paper airplanes and the runway to explain why forward speed matters.

Winner: Meta wins. It kept the energy high, matched a 10-year-old's reading level without talking down to them and delivered a clear, memorable takeaway.

5. Creative reasoning

Prompt: Invent a new app that solves a small but annoying everyday problem people have at home. It can't be a to-do list, meal planner, budgeting app or smart-home remote. Give it a name, explain the problem it solves, describe how it works and tell me why someone would actually keep using it after the first week.

ChatGPT solved a universal human frustration without requiring whole-house cataloging. The interaction model involves snapping a photo and speaking natural language ("Christmas lights in the blue bin") is low-effort.

Meta identified a hyper-specific, universally relatable domestic failure point and gave the app a brilliant, punchy name (Takeout). Its retention logic is rock-solid: instead of pretending users love opening chore apps, it acknowledged that the best utility apps require zero maintenance and rely on loss aversion.

Winner: Meta wins because the chatbot built a product around the reality of household laziness and recognized that an app survives past week one only when it demands almost no manual input from the user.

Final thoughts

After five prompts, Meta Muse takes the win 3-2. To say I'm both shocked and impressed is a serious understatement. While ChatGPT was stronger when a prompt required careful planning and attention to practical constraints, Meta Muse stood out when the task depended more on understanding people. Its neighbor text sounded like something I'd actually send, its airplane explanation was perfectly pitched for a 10-year-old and its app idea showed a surprisingly good understanding of why people abandon apps that require too much effort.

Meta Muse feels very much like it's own model, not one that's trying to imitate or catch up to ChatGPT. The little "Muse" at the top is unnecessary, but adorable.

ChatGPT is still my go-to, but this test proves that Meta has come a long way and this latest model offers a pretty strong first impression.

More from Tom's Guide

• Nvidia says 'local AI is here' — and it could change how you use AI at home
• OpenAI is paying influencers to convince you to use ChatGPT — here’s what we know
• I tested ChatGPT-6 vs Claude Opus 5.5 with 5 everyday prompts — it wasn’t even close

The Meta VR Glasses, which won't launch until spring 2027 for $1,299, were announced at an interesting time, just as Google and Xreal's collaborative Project Aura is expected to launch before the end of 2026 — there's no confirmed price for Xreal's device yet.

Both are spatial computing virtual reality platforms in a pair of smart glasses that are tethered to a puck that holds the battery and CPU. Similarities abound between the Meta and Xreal glasses, from the same Qualcomm silicon to somewhat similar gesture controls.

However, there are also plenty of differences, from the design and operating system to the AI features. Here's how these two wearable spatial computers are stacking up so far.

Meta VR Glasses vs Xreal Aura: Specs

Where Meta VR Glasses win

• A better, clearer display with no edge blurring
• Meta has experience with VR and tracking and it shows
• A more unique and interesting design

Unlike Snap Specs, what Meta and Xreal are offering between the VR Glasses and Aura are nearly identical experiences. This is scrunched-down virtual reality in a split-device experience with a compute puck and the tethered glasses.

But there are some clear differences, and one win for Meta is that its display is super-crisp with no edge blurring — unlike the Xreal Aura, which I found does have some.

Meta has touted that its glasses are Dolby Vision, Dolby Color and IMAX-certified, and you can certainly feel that work has been put into making these glasses hi-def when you're watching movies and 3D content.

The experience of using the VR glasses is pretty great thanks to near one-to-one tracking of not only your hands but also your eyes. The VR glasses feature multiple sensors in the lenses that let you use your eyes to select icons just by looking at them. You can feel Meta's VR experience coming to the fore here, something Xreal lacks as an AR glasses maker.

Plus, the exterior sensors can see your hands even if they're resting in your lap. The Auras still require that awkward reaching into the nothing to pinch at an invisible ball or whatever. With the Meta, you can lazily just keep your hands on a table and pinch your fingers or swipe your thumb when you're ready to use them.

In the world of AR glasses, Xreal's Aura wouldn't stand out, they look much the same as the Xreal One Pro, Viture Beast, or RayNeo Air Pro 4. But in this new spatial computing world, no design is the same, from the Top Gun-esque VR Glasses to Snap Specs to the Aura.

Personally, I find the Meta glasses more interesting-looking, even if they don't pass the "pub test" where you could wear them to a bar without getting comments. Xreal might let you get away with it.

Where Meta VR Glasses lose

• It's not quite VR, not quite smart glasses
• $1299 is expensive, even compared to Meta's own devices
• Field of view will bother some

I've been on record as finding VR headsets bulky and isolating, and personally I think the slimmed-down Meta VR Glasses are great with peripheral views and pass-through on the lenses.

However, for those who like the immersion of VR headsets, these probably won't do it for you. I've already seen complaints about the FOV and immersion online. As one reader said about our hands-on "They are still too heavy and look silly and having a wire connected to an anchor is a mom (sic) starter."

However, for smart glasses, the 70-degree field of view (on both Meta and Xreal's specs) makes it the largest of any existing AR glasses. I think it's great, but again, VR enthusiasts might be disappointed.

Where the Meta VR Glasses might come up short is the price. At $1,299, that's twice the cost of the Meta Quest 3 and three times more than the Quest 3S. That cost is more equal to the Steam Frame's high-end 1TB version. And that price doesn't include controllers if you want to enjoy VR games, though Meta promises more than 75 games that are "hand-enhanced" will launch with the glasses next spring.

It's more doable than the $2,195 Snap Specs, but $1,300 is still a hefty price.

Where Xreal Aura win

• Software and AI advantage
• Connectivity
• Presence in the world

During Meta Connect 2026, I heard some people who tried out the VR Glasses as having stolen its UI straight from the Apple Vision Pro. Honestly, it is very similar.

However, the advantage Xreal has is its partnership with Google, meaning you get Google software and Android operating systems to play in. This means Android developers will be able to build more apps and experiences for the Aura. Yes, Meta does have plenty of VR devs that it can pull from for games but that's just games.

Plus, Gemini, at this point, appears to be a more mature AI platform compared to Meta's Muse AI. Gemini Live feels like real-time advice in the real world and not just a camera-driven guide à la the Meta Ray-Bans.

I was told that Meta VR Glasses can connect to other devices, but I didn't experience that at Connect other than brief multi-window spatial computing connection to a laptop. However, Xreal has spent years making its AR Glasses work with other devices like the Steam Deck, or as we found at AWE 2026, the Nintendo Switch 2, something no other spectacles can do.

Right now, Xreal has the advantage.

Where Xreal Aura lose

• Probably as expensive
• Worse gesture controls
• It's another device that isn't a replacement

Xreal and Google have yet to release a price for the Auras, but based on Meta's pricing and compared to Xreal's current offerings, I'm expecting something close to, if not exceeding, $1,000 for the Auras.

And at that price, you're paying for worse gesture controls than what Meta offers. In the few hands-on sessions I've had, the Aura hand gesture controls worked mostly smoothly but could at times lag or not pick up the pinching motions needed.

With the Meta VR Glasses, on the other hand, I had no problems with the eye or hand tracking (though the UI highlighting needs some work).

Finally, and this goes for both devices, the Auras and the VR Glasses are another device I'm not sure people have room for in their lives. Yes, they can be used for gaming or as a monitor extension. But they both have the powerful Snapdragon Reality Elite chipset in the pocket and can work as computers or consoles, more or less. But would you?

I find both devices interesting, but I'm not sure you'd wear these in public outside of flights or the train. And if you're at home, you'll just use a console or your computer, especially with the 3-hour battery life.

Outlook

With both devices unreleased, there's still a lot we don't know about how they will work and what's changed since our hands-on sessions. Developers at Meta Connect seemed to be specifically seeking feedback. I expect some of the software problems I had will be altered between now and next year.

Still, both the VR Glasses and Xreal Aura promise a big, fascinating step into the future of spatial computing by squeezing VR into AR glasses infused with AI and a standalone OS.

What do you think about the Xreal Auras and the Meta VR Glasses,? Let us know in the comments.

More from Tom's Guide

• Ray-Ban Meta Audio hands-on — an attempt to allay privacy concerns
• What was missing from Meta Connect 2026? An apology
• No longer science fiction: How RayNeo AR glasses change everyday life and entertainment

Not only has it surpassed ChatGPT on the charts in Apple’s App Store, but it’s also racked up half a million downloads. Originally, I played around with the desktop version of Meta’s AI assistant and could see the potential for it becoming someone’s go-to chatbot. Now that it’s emerged as one of the more popular mobile AI apps, I downloaded it to take it for a spin.

Since I’m always in the market for comic books, video game controllers and physical games themselves, I came to Muse (where I've nicknamed my avatar “Amani”) with three goals in mind: find the best deals on three specific items from those aforementioned product groups.

After asking Meta’s AI tool for the proper prompts to achieve those goals, I hit the jackpot after using three of them to find what I was looking for.

Meta’s recommended deal-finding prompts for Muse

To get Muse started on its mission of helping me nab a few things as cheaply as possible, I presented it with this simple request: “What are the 7 best prompts I can use with you to help me with shopping to find the best deals?” It went ahead and listed the following deal-hunting prompts that are bound to get the best results out of it:

• Baseline + deal hunt: Finds the normal price first, then hunts for anything selling below it. Prompt: Find me a deal on [product]. First tell me the typical price, then find listings selling below it.
• Coupon and promo code sweep: I search for active coupon codes and promos for a specific product or store. Prompt: Find active coupon or promo codes for [product or store] and tell me which ones actually work right now.
• Compare across stores: I check the same item at several retailers and rank by total price. Prompt: Compare prices for [exact product name] across major retailers and show me the cheapest in-stock option including shipping.
• Set a max budget: Best when you know what you want to spend; I filter everything above it. Prompt: Find the best [product] under $[amount]. Prioritize quality and well-known sellers.
• Open-box/refurbished/used: Big savings live here. I bias toward resale and refurbished listings. Prompt: Find me a used or refurbished [product] in good condition. Max budget $[amount].
• Facebook Marketplace local deals: Great for furniture, bikes, and gear near you. Prompt: Search Facebook Marketplace for [item] near me under $[amount], local pickup only.
• Price-drop watch: I check back on a product you're eyeing and tell you when it dips. Prompt: Track the price of [product + link if you have it] and let me know if it drops below $[amount].
• Quick tip: The more specific you are (exact model, size, color, budget), the better the deals I surface. Want to try one now?

Finding the items I’ve desperately wanted with three of those prompts

Sadly, my search for active coupons through Amazon, a max budget search for a cheaper DualSense Edge wireless controller, Facebook Marketplace local deals for a sofa and a price drop-watch for an Xbox Series S either didn’t yield the most trustworthy results or produced nothing at all.

Thankfully, I did get back some inspiring options after testing out the prompts catered toward the “baseline + deal hunt,” “compare across stores” and “open-box/refurbished/used deal hunting commands.”

Entering all of those prompts into Muse kicks it right into gear as the AI scrapes all the available options across every retail site it can think of for whatever I’m on the hunt for.

The baseline + deal hunt prompt actually came in handy when I gave it this mission: find me the cheapest option for the Batman by Jeph Loeb & Tim Sale Omnibus.

Muse’s thorough search hit me back with this encouraging message:

BATMAN BY JEPH LOEB AND TIM SALE OMNIBUS HC at $74.99, new and in stock. That’s a 40% off the $125 list price, and a hair under InStockTrades’ $75 street price. For context on the rest of the field: MyComicShop’s reprint was $95, the cheapest new eBay listings were $79.99, and Amazon wouldn’t cooperate with the automated check.

I ended up heading straight to CheapGraphicNovels to get that hardcover gem, thanks to Muse’s best listed price.

The compare across stores prompt came in extremely clutch when I was in the market for a Midnight Black PS5 DualSense controller. The cheapest one Muse found was sold through Walmart with a comfortable price of $67.61 and free shipping, which felt like a steal compared to the controller’s standard price of $74.

The fact that it was being sold on Walmart through a third-party marketplace seller was no issue to me, especially since that seller had good reviews for the other DualSense controller it sold.

To wrap up my deal-finding sessions with Muse, I tested it with a prompt focused on open-box/refurbished/used deal hunting commands.

This time, I crossed my fingers in hopes of nabbing a physical copy of the PS4 version of Garou: Mark of the Wolves. After digitally perusing through a bunch of affordable options via eBay and Rancho Retro Games, Muse succeeded once again by finding me a used copy on eBay (rated as “Very Good”) priced at $54.99 from a seller with a 100% positive feedback rating and over 600 successful sales.

Final thoughts

The emergence of Meta's Muse as one of the best AI apps was a surprise to me, considering all the backlash being brought against Meta as a whole as of late. However, after giving it a fair chance, I walked away with a positive outlook on Meta’s AI agent and how it could lead me down the right path to save on all sorts of products I shop for online.

I’m sure Muse is a lifesaver for those who want to tap into it for productivity. For now though, I’m just grateful to have another AI tool at the ready to help me with my upcoming holiday shopping.

More from Tom’s Guide

• I replaced ChatGPT with Siri AI for a day — here's what surprised me
• OpenAI contractors were reportedly fired for using AI to train ChatGPT — here’s how they got caught
• AI is changing work faster than we can adapt — this futurist has a 30-day plan for workers to keep up

Most of the responses debated whether or not these glasses even count as VR or seemed excited to try out Meta's new product.

"Can’t wait til they are more affordable. I have a quest 3 and I use it mainly for exercise via fitness apps. But I have neck issues and I have been patiently waiting for headsets to get less bulky and lighter," one commentor said about the glasses and their $1,299 launch price.

Field of View

Of the comments we received that were negative the majority were focused on light leakage and the 70-degree Field of View.

"Looks like two steps back for VR to me. 70 fov may be enough for AR, not for VR," Mario Basiola said. "All that light leakage may be good enough for AR, for VR it breaks immersion. These are AR glasses, not VR."

"VR doesn't have to be all encompassing like we are used to," one commenter responded.

The concern is that 70 degrees is far less than the 110 degrees on the Meta Quest 3, leading to a less immersive environment.

VR headsets also tend to fully block out your surroundings, which does help make games and movies more immersive.

Meta's VR Glasses intentionally have peripheral gaps, meaning you should see light and your surroundings from the base of the glasses and the sides. I was told by a spokesperson that this was done on purpose after surveying VR headset users who wanted some peripheral vision.

Our reader's don't appear to agree with that. "It looks a bad idea for anything requiring immersion. The light leakage will be horrendous."

The price

While not everyone cares about the FOV or light leakage, even the most positive comments are cautiously excited but have doubts about the price.

"This is a huge step indeed, but it has to get cheaper, not more expensive. Now if you want to keep gaming, you have to be rich or make choices again."

And yes, the $1,299 price is high, especially compared to the $599 Quest 3 VR headset. And its apparent base price is higher than the $1,059 for the Steam Frame, though the 1TB variant of Valve's headset costs $1,299 as well.

Many of these comments came from the Tom's Guide Facebook page. Give us a follow there if you use that social platform. If not, you can also find us on Instagram, TikTok, X , YouTube and Reddit.

Also, feel free to let us know what you think about the Meta VR Glasses in the comments below!

More from Tom's Guide

• What was missing from Meta Connect 2026? An apology
• Ray-Ban Meta Audio hands-on — an attempt to allay privacy concerns
• Ray-Ban Meta (gen 3) smart glasses — lots more styles, longer battery life, but same privacy controls

An apology.

This entire year, there have been dozens upon dozens of reports of smart glasses being used to record and harass individuals (many of them women), with those recordings appearing on Instagram and other social media platforms, where they’ve been amplified.

There was an entire cottage industry of people who you could pay to remove or disable the recording light in Meta smart glasses, so others wouldn’t know if you were filming them.

Over the summer, the backlash against “pervert glasses” grew so strong that DuckDuckGo, the privacy browser, cheekily released its own “smart glasses” — just regular sunglasses, no tech whatsoever.

A research paper published in August from the University of Sydney highlighted how Instagram influencers regularly used Meta smart glasses to harass women in public or semi-public settings.

Yes, Meta did update the firmware in its glasses to prevent them from recording video if the light was blocked or disabled, and it did ban a number of Instagram accounts. But you would think that a company whose reputation was so damaged that its products were now called “pervert glasses” would do a little more in the way of damage control when it had the spotlight all to itself.

The backlash against smart glasses is only growing stronger. Last week, several local governments in Australia voted to ban them from even more public spaces while restaurants, pubs and theaters in London are also considering similar bans.

Meta’s new audio-only smart glasses can be seen as a response to this backlash, but it’s not enough. And, as Zuckerberg himself said in an interview with Joanna Stern, it’s not even that. “We didn’t decide to design this because people had some questions two months ago,” he said.

In the same interview, Zuckerberg said that Meta has added privacy features to its glasses over time, but should have also done a better job in advertising them.

Of course, as Stern noted, it took her reporting to alert Meta to the fact that people were modifying its glasses. Zuckerberg noted that the number of glasses that were bricked as a result of their recording lights being tampered with amounted to “less than one tenth of a percent,” which would amount to several thousand glasses.

During the keynote, Zuckerberg merely reiterated the protections already in place. “We built several leading privacy features for glasses from the beginning as well, including the light that turns on any time you take and save a photo or video,” he said, “and it disables capturing images and videos if that light gets tampered with.”

Meta's CEO then went on to talk about privacy protections as it relates to Meta being able to see how you interact with its Muse AI, but nothing more about preventing bad behavior of those using those devices.

Smart glasses are still growing

Maybe Zuckerberg doesn’t feel he needs to apologize for anything. After all, smart glasses sales are booming. Counterpoint Research released a report that said shipments in the first half of 2016 grew 263% year over year, and that Meta accounted for 94% of it. That’s a lot — in 2025, EssilorLuxottica, the company that actually makes Meta’s glasses, said that it sold 7 million pairs, up from 2 million in 2023 and 2024 combined.

An IDC report predicts that smart glasses will grow 35% to 13 million this year, and nearly double to 25 million by 2030."

But, when even ICE bans its officers from using smart glasses, there’s more than an idle concern over the intrusiveness of these wearables.

It’s a shame. Smart glasses with cameras do have some real benefits, from helping know what’s around you, to aiding those with vision problems. But having something with a camera that’s constantly recording — or is capable of doing so — introduces a host of privacy issues that Meta didn’t consider, and even now, only seems to be paying lip service to.

When a very unfinished Apple Maps launched, Tim Cook issued an apology. This past summer, when it was revealed that law enforcement officials were abusing its technology, Flock’s CEO apologized. Even Mark Zuckerberg once apologized to the parents of children who were victims of online abuse, though it took a senate hearing to do so.

Even an acknowledgement that Meta's glasses were being used for ill intent, and some note of sympathy for the victims, would have been at least something.

Short of being called up to Capitol Hill, what’s it going to take for Meta to show any sort of contrition? Yes, the company itself wasn’t the one harassing women, but its tools and technology enabled the behavior, and it didn’t do enough to stop it sooner.

More from Tom's Guide

• I'm worried 'pervert glasses' are going to ruin all smart glasses
• Less pervy? Meta AI glasses will now stop recording if you cover the warning light
• I just went hands-on with the new Samsung smart glasses

Meta’s recently released AI companion app, Muse from Meta, has garnered a ton of attention—in fact, its popularity has resulted in it taking the #1 spot in the “Top Free Apps” category in the Apple App Store and overtaking its closest rivals ChatGPT, Google Gemini and Claude by Anthropic. Now, the company most known for its association with Facebook, Instagram, WhatsApp, Meta Glasses and Meta VR headsets is taking another stab at diving into the physical product market.

While on stage at this year’s Meta Connect conference in Menlo Park, California, Zuckerberg made a major announcement: Meta Muse AI users will get the chance to interact with their new favorite digital assistant through an upcoming device. And to top it all off, it’s coming a lot sooner than we thought.

Say hello to the Muse Charm.

Meta’s Muse Charm is going the Tamagotchi route with its presentation and practical uses

The Muse Charm device was unveiled onstage during Zuckerberg’s Meta Connect presentation. Those in attendance and watching at home were treated to a pocket-sized, keychain-attachable piece of hardware that will give its users a direct connection to their Muse AI assistant without having to access it through the app on their phones.

Just from taking a look at it, the Muse Charm looks to be roughly the same size as your average AirPods case. In a move that should give avid fans of Tamagotchi a wave of nostalgia, the Muse Charm places its animated avatar on a miniature 2-inch OLED touchscreen that can be activated by simply tapping the screen to activate its fingerprint sensor and talking to it. The device’s other features include speakers, microphones, front- and rear-facing cameras and built-in 5G connectivity.

The camera integration stands out as one of the Muse Charm’s most impressive features, as the AI assistant inside it will be able to see what you’re pointing it at and be able to identify products, read signs and analyze assorted real-world items in real time.

Among the many tasks the Muse Charm is capable of handling are sending emails, booking travel, organizing tasks, developing plans, coordinating resources, turning saved Instagram recipes into grocery lists, planning dinner parties, remembering relevant preferences and interacting with connected services. Speaking of connected services, Meta’s Muse AI agent has partnered with several prominent companies, including Google (Gmail & Drive), Shopify, OpenTable, Ticketmaster, PayPal, Lync, and many more. This X thread showcases the full gamut of Muse AI connected services that should work in conjunction with the Muse Charm once it releases.

The Muse Charm is said to begin shipping sometime in December, just in time for the holiday season. No details about its official price, battery life or complete hardware specifications have been announced yet.

Zuckerberg’s comments from his Meta Connect presentation point to Meta recognizing the need to offer physical products that give users a frictionless way to interact with AI. "It is actually quite clear personal superintelligence will not only come ⁠first; it will be much bigger and more important," Zuckerberg stated. "So we have shifted our focus to building the best devices ​for personal superintelligence," he said, adding that Muse would also soon be available across Meta's smart glasses lineup.

The takeaway

I remember when similar AI companion devices like Humane's AI Pin and the Rabbit R1 drew negative reactions. Will Meta’s Muse Charm be any different from those previous failed attempts? Fingers crossed that Meta learns from its miniature AI device predecessors' errors and emerges with another major winner in the AI space. With a connection to Meta's broader AI, social, messaging and hardware ecosystem right out of the gate, the Muse Charm may just have everything it needs to succeed.

More from Tom’s Guide

• I used Siri AI to plan my entire weekend — and it did way more than I expected
• I tested ChatGPT-6 vs Claude Opus 5.5 with 5 everyday prompts — it wasn’t even close
• I tried Apple’s new at-home wellness assessment in iOS 27 — and gained valuable insights into my flexibility and balance

You can imagine the skepticism I wore on my sleeve when I was given a chance to try out the new Meta VR Glasses during Meta Connect 2026. And my pleasant surprise at how intuitive and comfortable it was to wear and use the headset/glasses.

Meta's newest VR wearable is a cross between the Meta Quest headset and AR glasses like the Viture Beast or the upcoming Xreal Aura. Or if you're just looking at gesture controls, a much cheaper version of the Snap Specs (but still expensive).

While the race is on to see who can dominate the AI glasses side of things, Meta's VR glasses feel like the actual future. As a VR headset they're still isolating, and there are some strange decisions under the hood, but there's something interesting here.

What are the Meta VR glasses

Meta's VR Glasses are what happens when you slice off just the lenses of a headset and put them in a pair of Top Gun fighter pilot goggles. The glasses carry the lenses, display and sensors.

The lenses are a pair of Micro OLED 5K "ultra compact" pancake screens that Meta says are Dolby Vision certified and have Dolby color. It's also capable of running 3D content and is IMAX certified, a claimed first for headsets.

The chip and battery have been separated into a tethered puck that Meta says can last for three hours on its own. A Meta spokesperson told me the glasses feature a Qualcomm XR2 Gen 3 chipset. Or, by another name, the Snapdragon Reality Elite, announced at AWE this year. The naming convention is confusing since Qualcomm didn't announce XR2 Gen 3 but instead the rebranded Reality Elite.

Seriously, these things are light. Meta Quest headsets weigh 1.1 pounds or about 500 grams. The Meta VR Glasses are supposed to weigh only 100 grams or 3.57 ounces. For comparison, the Viture Pro and Xreal One AR glasses weigh just under 3 ounces.

They're quite comfortable to wear and don't get hot thanks to the puck. However, the puck was quite warm when I picked it up at the end of the session.

Perhaps the most interesting thing about the VR glasses is the lack of controllers. They can support controllers like the Touch controllers used with the Quest via Bluetooth. However, these glasses utilize eye and gesture controls for most every interaction; more on that in a moment.

Price and release date

We have some time before we'll get our hands on the Meta VR Glasses for real. They're not expected to ship until spring 2027 — six months, I was repeatedly told.

At launch, the Glasses are supposed to have a $1,299 starting price, though no storage capacity options were given. Meta Quest 3 is only offered with 512GB, while the 3S comes in 128GB and 256GB versions. I presume the VR Glasses will probably launch at 256GB.

The price is high — they're more than twice as much as the Viture Beast — but about half the cost of the Apple Vision Pro headset.

All about the eyes

With the caveat that these were locked-in demos, the eye-tracking and hand gesture controls felt intuitive and worked extremely smoothly.

One cool thing I found is that the glasses will walk you through setting up the lenses and nose pads for your vision and nose shape. With my eyes, I had to adjust the right lens, which was done by just pushing it to the left a slight bit.

For prescription glasses users, Meta had a prescription setup so they could calibrate the lenses. It's not clear how that would be done for customers once it's in the wild.

One demo showed off a multi-screen workstation where what caught my eye was the readability of the websites and code I was shown. It fixes one annoying flaw of AR glasses where reading can be blurry or straining on your eyes.

There are sensors in the lenses that track your eyes and do a fine job of telling where I'm looking. I was able to look at an icon and then pinch my fingers, and the app would open.

Nicely, unlike many devices with gesture controls, I could keep my hands in my lap or flat on the table if I wanted. There's no strange reaching into the aether for an invisible ball that you usually see.

This is thanks to sensors on both the front of the glasses as well as the underside that capture your hand movements.

Content, content, content

While the Meta Quest or Valve's new Steam Frame function as gaming headsets, Meta stayed away from gaming during presentations about the headset. Yes, there are 75 launch titles that are "enhanced to be hands first," but this isn't strictly a gaming headset.

Much of what I was told and shown revolved around different content. There are the Dolby-certified lenses and IMAX certification. Meta will have a store with 3D movies and shows that will work with the glasses, along with new 3D titles coming from Lionsgate and James Cameron to start.

I glimpsed some of that with a Marvel Avengers scene where a 360 3D environment was built out around me while I watched Thor build his Mjolnir replacement axe. It was cool if a bit of a theme park ride. There was also a Billie Eilish concert where, using 3D it looked as though I was standing in the crowd, which felt creepy like I was breathing down someone's neck

There was also a 180 degree spatial version of the NBA Finals where I could turn my head and see Timothée Chalamet watch the Knicks beat the Spurs. This is supposed to come with more sports, including ESPN and college football that starts this October on the Quest.

On most AR glasses, a limited FOV determines the virtual screen size. With the VR glasses, I was not told what the FOV was, and for screen size I was only told I could expand or shrink the displays I was watching, though that option wasn't available during my hands-on session.

Gestures need work

During my hands-on time, it was tough for me to get the hang of using both eye tracking and virtual hand gestures.

I found it quite difficult in the Meta UI to tell if I was selecting something with my eyes. When I tried opening the Billie Eilish concert, I had no clue if the play button was actually selected or not.

This isn't a tracking issue; it's a highlighting problem. Instead of adding a bold border or changing colors, the icons you select just get brighter, like you've shone a penlight on them. On some darker shaded apps this is fine but on white play buttons it was nearly indistinguishable.

Presumably this could and should be fixed before launch.

The other problem I had was with the touted "Surface keyboard." It's a virtual keyboard and trackpad that you can put on a table in front of you and type. It looks cool, but in practice it's hard to actually type on without any haptic feedback on what your fingers are doing.

I mentioned using haptic devices like gloves or VR gaming vests like those from bHaptic, and Meta spokespeople didn't seem to know if they would work or thought it would confuse the headset. Pour one out for your VR headset accessories.

My verdict: The future of VR headsets

Right now, Meta and its Quest headsets dominate the VR headset market. Per Counterpoint Research, Meta has nearly 70% of the market. And if we're being honest, VR is just the Quest right now, though the Steam Frame could change that conversation a bit.

As mentioned, I find VR headsets bulky and isolating. I don't know that Meta VR glasses solve the second issue, but bulk is nearly gone, and it makes for a fascinating device.

This new form of VR headset is so intriguing and an actual improvement over current VR platforms that I believe it's only the start, and I hope, just the first step in a fascinating new direction for virtual reality devices.

We're slowly getting to a point where AI glasses and VR/AR headsets and glasses are going to become one and the same. This midpoint evolution shows both how far away we are and how far it's come.

More from Tom's Guide

• Vonder may have solved the 'pervert glasses' problem — and the nerd factor, too
• I bought Instagram's new Meta One subscription — here's what $7.99 actually gets you
• Meta just launched an AI that can actually run your life online — here's what Muse can do

The standard Ray-Ban AI glasses are slightly thinner, with a more comfortable fit via more adjustable temple tips. The big update is more styles, colorways, and lens combinations — but no major changes to the company's privacy policies.

Two new styles

Now, Ray-Bans can be purchased in 27 color and lens combinations, including Black, Havana, and select seasonal colors across three styles. Plus, there are two new frame styles and a larger version of the Wayfarer frames.

The newest style is the Ray-Ban Meta Aviator frames, which copy the famous Aviator style, à la Top Gun. The second new frame is the Ray-Ban Meta Zena, which is a cat-eye frame.

Price and availability

Ray-Ban Meta (Gen 3) is available today starting at $449.

Also upcoming is the new hearing-enhanced feature that will launch on Ray-Ban Meta glasses first. This feature is supposed to support people with mild to moderate hearing loss. This feature is supposed to isolate sound and help you hear better. It will cost a one-time $149.99 fee or be included as part of the Meta One subscription. A specific launch date was not announced.

Longer battery life, more mics

The third generation gets nine hours of battery life, one more than Gen 2, and a new 6-microphone array that is meant to cut down on background noise.

The new action button is customizable but is mostly meant to give you instant access to Meta AI.

Meta is also adding the ability to take landscape- or portrait-oriented photos and dynamic photos. Dynamic photos take a slew of photos that you can pick from when you try to capture a moment or a group that won't stop closing their eyes.

Meta spokespeople said that later this year a software update will let you capture videos in Dolby Atmos and capture sound in 360-degree spatial audio.

Outlook

And that's about it. If you're looking for more styles and colors, then Meta has you covered, but if you're hoping for more updates, 2026 is fairly quiet.

We were hoping that Meta would announce new privacy controls for its glasses, but those appear to have been rolled out during the summer. However, the company did announce a new audio-only pair of Meta glasses, as well as a new VR headset with hands-free controls.

More from Tom's Guide

• I’ve traveled over 50,000 miles testing smart glasses in 2026 — here are my 3 favorites that have launched this year so far
• I tested Google’s “Intelligent Eyewear,” and found the smart glasses that will defeat Ray-Ban Meta
• Tom's Guide to smart glasses: 13 terms you need to know

Plenty of other companies, like Even Realities, whose CEO decried cameras on glasses over the summer. "Adding a camera onto a pair of glasses definitely introduces many more downsides than upsides."

With those concerns in mind, I tried out the audio-only Meta glasses during Meta Connect 2026. Will Meta's new camera-free specs open up a new segment for the company?

They're Ray-Bans

On the surface, these glasses are just Ray-Ban Metas. The lack of a camera means that they're lighter and enables Meta and Ray-Ban to introduce new styles like the Clubmaster, which features more of a wire frame around the lenses.

The glasses weigh just 43 grams, which is roughly 10 ounces lighter than the Ray-Ban Meta gen 2 specs..

They do have more battery life than the camera-version, hitting a claimed 12 hours over the recently updated 9 hours in the Gen 3 version of the Ray-Ban Meta glasses.

Per Meta, the glasses come in 23 color and lens combinations between the Clubmaster and Burbank styles.

The glasses launch on October 13 for $349. The price is perhaps the most surprising thing about the Audio glasses. Removing the camera apparently does not make them cheaper.

Yes, the Ray-Ban Meta (Gen 3) glasses start at $449, but the currently available Gen 2 glasses start at $379, only $30 more expensive. And the still-available Gen 1 versions start at $224.

Similar uses

Using the Audio glasses is exactly the same as using the camera-laden version.

Swipe on the right arm for volume controls. Tap the button to access Meta AI or say "Hey Meta" before prompting the glasses.

If you've ever tried a pair of Ray-Ban Meta glasses, you've tried the Audio glasses, sans taking pictures or videos. Both the Audio and Gen 3 Ray-Bans come with six microphones for taking calls and talking to the AI assistant. It also comes with speakers for music and podcast listening.

The speakers sound about the same as on-camera Ray-Bans, which is decent for what it does.

I've received conflicting information on whether or not the new Hearing Enhancement feature for "mild to moderate hearing loss" will be available on the Audio glasses when it launches. At one point, I was told it would launch on the Gen 3 glasses first, but another spokesperson said it's coming to all Ray-Ban glasses. Press releases on the feature have not been illuminating.

Early verdict: Ray-Ban Meta Audio

On the whole, if you enjoy your Ray-Bans and don't really use the cameras or don't want to be seen wearing the camera-version, the Ray-Ban Meta Audio glasses are nice option to have.

They're fairly comfortable, the user experience is the exact same and they look great, especially the Clubmaster frames.

The longer battery life is nice and basically guarantees nearly all-day use for the Audio glasses, something the Gen 3 glasses can't promise.

However, the price is high for glasses that offer fewer features than the normal ones, especially when the older versions of the Ray-Ban Meta glasses are still for sale for cheaper or about the same.

More from Tom's Guide

• OpenAI's new ad tracker may know what you do after you leave ChatGPT — here's what we know
• The guy who predicted the housing crisis is calling B.S. on the AI apocalypse — here’s why
• I bought Instagram's new Meta One subscription — here's what $7.99 actually gets you

So at the Snapdragon Summit, I got to talk to Ziad Asghar, Qualcomm’s SVP and GM of XR about it, and he mounted a passionate defense of the smart glasses category.

But the issue is clear: while Qualcomm’s hardware-level privacy tools are genuinely impressive, they rely on a big leap of faith regarding consumer behavior and corporate accountability.

“It’s just a phone”

When I pressed him about the public backlash against face-mounted cameras, Asghar brushed off the anxiety as a generational growing pain.

“Today, you carry devices that have multiple cameras on them. All of you do. It’s called a phone, right?” Asghar commented. “I think different age groups are thinking about it in a very, very different way, because the younger people are so used to people blogging and vlogging and carrying their cameras in public spaces and all. So I think some of the concern honestly, in my opinion, is just because it's a new form factor.”

Fair answer, and I’ll give props to the silicon-level tools Qualcomm is building in a minute. But here’s where I have an issue with that — a smartphone and smart glasses are not socially equivalent. Lifting a rectangular slab and aiming it at someone is an overt, highly visible social cue.

However, wearing sunglasses that look near-identical to Ray-Bans while recording is inherently covert. The “new form factor” isn’t just a novelty; the physical invisibility of the camera is the privacy threat.

The genuine breakthrough

If there is a saving grace in this debate, it is the actual silicon Qualcomm has built. Asghar detailed how Snapdragon chips are engineered to process privacy on-device, completely bypassing the cloud. The most critical tool is hardware-level face obfuscation.

"On our always-sensing capability on our chips, it can detect a face — not recognize, but just detect a face,” Asghar highlighted. “For multimodal AI, I don't really need to send all these faces for AI processing. What I really need to do is to be able to send the context: 'Oh, I'm sitting in a conference room over here, it's a work-like setting, it's not my home.' For that, I can detect the face and I can cover it up. That technology is something that we have integrated into our products and we are sharing with our partners."

This is the exact technical safeguard the industry needs. By using the NPU to censor faces before the video frame is handed off to an AI agent, Qualcomm can guarantee that bystanders’ biometric likenesses are never processed or sent to remote servers.

The OEM loophole: Qualcomm suggests, Meta decides

But the fatal flaw in Qualcomm’s privacy shield is that they do not control the final product. Qualcomm builds the safety net, but device manufacturers (OEMs) hold the keys.

"We provide the enabling technologies," Asghar explained. "But the end device experience is developed by our end customer. So far across the board, all of my glass customers today have LEDs. You take a picture, the LED comes on, tells the person in front of you, 'Oh, you just took a picture.'"

Qualcomm can build chip-level face blurring and tamper-resistant camera pipes, but it’s tricky to force companies to turn them on. If a hardware partner decides that local obfuscation slows down their visual search latency, or if they actively want raw visual data to train their own models, there’s nothing stopping them from bypassing the local protections.

Will consumers punish bad actors?

To wave away the OEM loophole, Asghar argued that brutal market economics will force hardware makers to prioritize privacy.

"If they launch a product and there are doubts about it, the product may not do very well, right? So it is in the interest of our customer to actually work with us,” Asghar responded.

However, history has proved otherwise. Sleek design, low friction and convenience has always won out over stringent data privacy in the eyes of many. And that’s where the problem lies.

I’ll give kudos to Qualcomm here. The company is absolutely building the right tools for a privacy-first spatial computing era. But providing the hammer doesn’t mean the carpenter will build a safe house.

Until these hardware-level privacy protections are mandatory, un-bypassable requirements for using these Snapdragon chips, the privacy of the public is always going to be questioned.

More from Tom's Guide

• I wore the new Snap Specs — the design is wildly polarizing, but the instinctual computing is a genuine breakthrough
• Snap’s CEO told me he hates getting AI-drafted texts from his wife — so what’s the plan to fight brain fry with Specs?
• I tested Snapdragon 8 Elite Extreme Gen 6 — 16% faster CPU and huge AI-infused gaming upgrades, but Apple’s A20 Pro pulls ahead in some areas

Meta Connect 2026 was kicked off by the company's annual keynote from CEO Mark Zuckerberg, which revealed what Meta has been working on.

As usual, hardware was a big focus of the keynote — Zuckerberg announced a new VR headset that could rival the Apple Vision Pro, a new pair of audio-only smart glasses, the Ray-Ban Meta (gen 3), and, in a real surprise, the Muse Charm, a small pendant-style device that contains the company's new Muse AI.

There was also a good amount of time spent on Muse and what it could do; while it had previously been revealed that it could make purchases on your behalf, at Meta Connect it was announced that it would work with a host of retailers, including Walmart and Instacart.

But that's not all — Meta also announced several other new features designed to enhance its smart glasses, including a new virtual environment to isolate your interactions with its Muse AI from the rest of Meta. However, there was very little said about the privacy issues that have dogged the company over the past year.

What was announced

Meta announced four new hardware devices at Meta Connect 2026

• Ray-Ban Meta (Gen 3) glasses — a refresh with longer battery life and a few new frame styles. Available today starting at $449.
• Ray-Ban Meta Audio - Meta's first pair of camera-free glasses. They will ship on October 13, and can be pre-ordered starting at $349.
• Meta VR Glasses - a new ultra thin VR headset aimed at entertainment and productivity. They will cost $1,299, and be available this spring.
• Muse Charm - AI device that holds an AI agent on your keychain. It will be available this fall, but pricing has not yet been announced.

In addition to hardware, Meta also announced a number of software features for its devices. Here are some of the highlights.

Muse AI

The company announced updates to its Muse AI, from integrations with numerous shopping sites, to the ability to fully customize your avatar, from its look to its name to how it speaks.

Hearing Enhancement

Available later this year on select smart glasses, this FDA-approved OTC software feature will be able to assist adults with mild to moderate hearing loss. However, this feature will cost $149 per year, or included with a Meta One subscription.

Improved navigation for Meta Ray-Ban Display

Meta's Display smart glasses will soon get cycling and public transit directions, as well as real-time transit departures. Available later this month in the US and Canada.

Hologram for video calls

When making a video call in WhatsApp using the Meta Ray-Ban Display, the caller on the other end will see a holographic model of your face, which will move in response to you talking. To use the feature, you use Muse AI to capture your face; this feature will roll out in Early Access in the U.S. later this fall.

Meta Connect 2026: time, date, and how to watch

The Meta Connect keynote took place at 4 p.m. PT (7 p.m. ET / 12 a.m. BST) and was livestreamed via Meta.com. It is now also available to stream on YouTube.

You can read through our full recap of the event below, including analysis from our two intrepid reporters on the ground at Meta's headquarters in Menlo Park, California.

RECAP: Full updates

As with the last few Connects, smart glasses should take up a bulk up of the presentation. However, this year, Meta is rolling into this event on a privacy backlash and the neverending RAM crisis.

Below, I've looked into rumors and leaks to see 6 major things that could get announced at Meta Connect 2026.

Three different smart glasses

Right now, Meta is facing serious concerns over privacy and it's "pervert glasses." One way the company is rumored to allay those concerns is by launching a new pair of camera-free smart glasses.

These glasses will reportedly have the audio features of current Meta Glasses. This means the glasses will have at least six microphones that let you ask for directions and do other AI-based tasks without taking your phone out of your pocket.

As for other glasses, we expect to see a gen 2 version of the Oakley Meta glasses and Ray-Ban Meta Gen 3. We know more about the Gen 3 Ray-Bans thanks to rumors. It should feature the latest Snapdragon AR chipset and new designs. New chips should mean longer battery life.

Both will probably get smart AI and improved audio.

A privacy interlude

Before we look at more devices and services, it will be interesting to see if Zuckerberg and co. address privacy during Connect.

In August, Meta strengthened tamper detection on its AI glasses to prevent people from blocking the LED recording light that indicates when the camera is working. The update makes the camera stop working if it discovers that the small white LED on the front of the glasses is covered.

"We're going to keep rolling out updates like this to make sure the capture LED can reliably alert bystanders when photos or videos are being captured for someone's gallery," Alex Himel, vice president of wearables at Meta, wrote on Threads at the time.

In the same month, Meta filed an AI facial recognition patent that would let its smart glasses recognize faces, detect laughter and record automatically.

One step forward, 15 back.

Project Phoenix

Meta has been coasting on the Meta Quest 3 and 3S headsets for a few years, but a VR leaker who goes by Luna on X found evidence of a mixed reality headset dubbed Project Phoenix in Meta code this month.

Luna's new images seem to confirm mock-ups that leaked in February of this year. At the time, it was rumored that the new headset wouldn't debut until early 2027, but Meta could still announce them at Connect.

Interestingly, based on the images that Luna dug up, the new headset looks somewhere in between a pair of AR glasses and VR headsets. The images of the Phoenix show an ear cable à la the Xreal Project Aura, which suggests the battery and silicon are probably on an external device.

It's rumored the headset will support hand and eye tracking. Otherwise, we don't have many specs, but it will likely feature the Snapdragon Reality Elite chip that Qualcomm announced at AWE 2026 earlier this year. That chip is specifically designed to support XR devices.

Muse, HorizonOS and "avatars"

Meta just launched Muse, its new Muse AI that can send emails, spend money and work when you aren't looking, two weeks ago. We expect Meta will spend some time highlighting the AI during Connect and how it works across devices.

As for its headset and VR operating system, Horizon, if Phoenix is announced, we should see some major updates to HorizonOS. Our friendly VR leaker Luna hinted that Meta could soon reveal its Hologram avatars that were teased with the announcement of the Project Orion AR glasses in 2024.

Last week, The Information reported that an avatar of Meta CTO Andrew Bosworth will make an appearance during Connect. It'll be interesting to see how Meta views its metaverse since its VR platform Horizon Worlds was shut down this summer.

How to watch Meta Connect

Tom's Guide will be attending Meta Connect, so keep us in your tabs to follow along with all the new announcements. The keynote speech starts at 4 pm Pacific (7pm ET) on Wednesday, September 23.

Meta will likely livestream the keynote on Instagram, Threads and Facebook. The company also shares it on YouTube if you prefer that platform.

More from Tom's Guide

• I bought Instagram's new Meta One subscription — here's what $7.99 actually gets you
• Meta just launched an AI that can actually run your life online — here's what Muse can do
• Meta agrees to pay $17.1 billion to settle claims Facebook and Instagram harm children — here's what you need to know

That means if you use ChatGPT and later visit a participating retailer or other website, OpenAI may be able to recognize the same browser and receive information about what you do there.

The finding comes at a particularly interesting time. OpenAI has been rapidly expanding advertising inside ChatGPT, recently announcing new AI-powered advertising tools and Sponsored Agents. The company said last month that ChatGPT Ads had already reached a $1 billion annualized revenue run rate.

But this isn't quite as simple as assuming ChatGPT is watching everything you do online. There are some important limitations to the research, including one major piece of the system the researcher wasn't able to independently verify.

Here's what we know.

What is __obi?

The research comes from an independent researcher who publishes under the name Buchodi and says they analyzed months of network traffic involving 936 OpenAI advertiser pixels across 1,029 hostnames.

According to the analysis, visiting ChatGPT can result in OpenAI placing a cookie called __obi in your browser.

Cookies themselves aren't unusual; you've probably accepted them on a several websites. They're small pieces of information stored by your browser and are used for everything from keeping you logged into a website to remembering preferences and measuring advertising.

What's interesting here is how this particular cookie is configured. Buchodi found that __obi can last for one year and uses the SameSite=None setting, which allows it to be included in certain requests originating from other websites.

That's where OpenAI's advertising system comes in. Businesses advertising through ChatGPT can install OpenAI's measurement pixel on their websites, similar to the tracking technology used by advertising platforms such as Google and Meta.

When Buchodi visited participating websites, requests sent back to OpenAI included the __obi identifier. In other words, the same identifier created through ChatGPT could show up again when that browser visited another website.

What can OpenAI see?

Potentially more than simply whether you visited a website. According to Buchodi's network analysis, OpenAI's advertising collector received information associated with pages people visited and conversion events on participating sites.

The researcher examined 23,929 observed URLs and found that query strings were stripped before being sent to OpenAI, but URL paths remained.

A URL path can sometimes reveal considerable information about what someone is looking at. Buchodi says paths observed in the dataset included pages involving a medical condition, debt services and a litigation intake form.

The advertising system also supports what OpenAI calls conversion measurement.

OpenAI's own documentation says advertisers can install a pixel and use conversion measurement to understand what happens after someone interacts with a ChatGPT ad. Its measurement integrations can support conversion tracking and attribution across websites, apps and even offline activity.

That's standard practice in digital advertising. What's unusual is that the advertising platform in this case sits alongside an AI assistant where users may share substantially more personal context than they would with a traditional search engine or social network.

Is it connected to your ChatGPT account?

This is where the story gets more complicated. Buchodi found that the process used to generate the identifier included tokens labeled either account_user or anonymous.

Of 932 tokens the researcher decoded, 736 were associated with logged-in users and 196 were labeled anonymous. The researcher also found that anonymous identifiers could remain stable on the same device for at least 27 days.

However, there's an important caveat. Buchodi did not directly observe OpenAI taking an event received from an advertiser website and joining it to a person's ChatGPT account on its servers.

The researcher argues that the architecture makes such a connection possible, but that final server-side step wasn't independently observed.

So it would be premature to say OpenAI definitively attaches every page you visit on participating sites to your ChatGPT profile. What the research does demonstrate is that the same OpenAI identifier can be transmitted from participating websites back to OpenAI.

Buchodi says the identifier appeared across a number of recognizable websites during testing. On one device, the same __obi value was sent to OpenAI from 12 commercial websites using 13 separate advertising pixels. Across the larger dataset, 12 of 30 identifiers appeared with more than one advertiser, while one appeared across 10 different advertisers.

The researcher also notes several important limitations. Testing was performed using Chrome on Android, and desktop Chrome wasn't tested. Safari's anti-tracking protections block the third-party cookie mechanism described in the research, meaning it shouldn't operate the same way on Safari or other iOS browsers using Apple's WebKit.

The identifier also wasn't generated during every ChatGPT session. Buchodi estimates that roughly one in five sessions tested produced the synchronization token.

What does OpenAI say?

OpenAI hasn't publicly responded to the specific findings. Buchodi says they contacted OpenAI's press and privacy teams on September 14 and asked why __obi was classified as an analytics cookie and how it behaves when someone rejects marketing cookies.

According to the researcher, OpenAI acknowledged the inquiry and said the findings would be passed along internally but had not provided answers to those questions when the research was published.

Tom's Guide has contacted OpenAI for comment and will update this story if we hear back.

OpenAI's public documentation emphasizes that advertisers do not receive ChatGPT conversations, chat histories, memories or personal details.

The company also says it does not sell user data to advertisers, and advertisers receive aggregated reporting about things such as views, clicks and conversions.

That's an important distinction here: the research does not suggest advertisers themselves can read your ChatGPT conversations or access the __obi identifier.

Instead, the concern is about information potentially flowing in the other direction — from websites using OpenAI's advertising tools back to OpenAI.

Why this matters

Cross-site advertising measurement isn't new. In fact, Google, Meta and other advertising companies have spent years building systems designed to determine whether someone who saw or clicked an ad later visited a website or bought something. But ChatGPT introduces another dimension.

People don't just search for products using ChatGPT. They use it to discuss relationships, health concerns, finances, careers, travel plans and countless other parts of their lives.

OpenAI itself promotes ChatGPT advertising around this idea, telling advertisers that people share "richer context" in ChatGPT while exploring options and making decisions.

That's why the boundary between your ChatGPT activity and what you do elsewhere on the web matters.

For now, the evidence doesn't show that OpenAI is handing your conversations to advertisers — and OpenAI explicitly says it doesn't.

But the discovery of __obi raises a different question: how much information should be able to flow back into an AI platform after you leave it?

What do you think? Share your thoughts in the comments.

More from Tom's Guide

• AI is turning ordinary photos of kids into sexual abuse imagery — here's how to protect your family
• OpenAI paid contractors to read ChatGPT conversations — here's how to protect yourself
• I stopped reading the fine print myself after discovering this overlooked ChatGPT trick

There was a small but healthy industry of individuals who would modify Meta’s glasses to remove or disable the recording light so you could video someone surreptitiously. And the ease with which wearers could capture content and then post it to Instagram — another Meta-owned property — has led to a rash of harassing videos being posted to social media.

In late August, Meta finally addressed some of these issues. It issued firmware updates to its glasses that prevent recording if the light is damaged, disabled, or blocked, and it banned a few creators and blocked several hashtags on Instagram that were widely used to disseminate the harassing videos.

But is this enough, or has Meta’s reputation been tarnished irreparably? To find out how we got here, and to see what Zuckerberg & Co. could do to revive the brand, I spoke to three analysts to hear their thoughts.

How we got here

Meta has been called out more than other smart glasses makers for the simple reason that it has sold more devices than anyone else — as much as 80%, according to one report. And those glasses lacked the privacy protections they needed.

“Meta is treading a fine line between those who abuse their glasses and those who do not,” Ramon Llamos, a wearables analyst for IDC, said. “Safeguards like the recording light haven't proven effective, and the stream of harassing videos has gone unabated even as several cases around unwanted recording have resulted in jail time.”

“The reason why Meta has been targeted is that they made the most popular glasses,” Carolina Milanesi, of The Heart of Tech, said. “People do not trust them, and they have not taken accountability. By doing so, they have made it much harder for everyone else.”

It’s not just Meta.

“There has definitely been a backlash against AI-enabled cameras, whether that is Flock used by law enforcement or smart glasses used to record people in public,” Avi Greengart of Techsponential said.

“However, Ray-Ban Meta sales have increased, so the challenge for Meta is to keep its momentum up heading into the holiday season: you want people to be delighted when they receive smart glasses as a gift, not horrified. To that end, Meta has several challenges: core use case, security architecture, and brand reputation.”

What Meta has to do

“No matter what additional steps Meta puts in place to improve privacy, Meta does need to fully explain its architecture and ensure that its glasses cannot record when they aren't worn, can't record without notification, and that captured footage can't be viewed by Meta or its contractors,” Greengart said.

Llamos said that Meta could make it more obvious that its glasses are recording someone. “I've heard of possibilities like playing back a sound to alert that the glasses are recording,” he said, “but those could penalize people who have been using the camera responsibly.”

Milanesi said that Meta “needs to go beyond just talking and show some progress on the hardware side, either with the light, making it even harder for people to disable, and investing in bringing solutions like a shutter to market.”

What other brands have to do — and are doing

Of course, the backlash against Meta’s glasses has other companies rethinking their strategies. At the launch of the Apple Watch 12, for instance, John Ternus was careful to highlight the safeguards his company put in place with all the new Siri recording capabilities.

“One way to avoid people being upset that you're recording video is to not record in a way that humans can ever view the footage, just use that data as context for AI, and ensure that the data is not used to train AI,” Greengart said.

“This is the approach Apple takes with its new Apple Watch summary tools. Meta can't do this — at least not fully — because taking hands-free, in-the-moment photos and videos is a core part of Ray-Ban Meta smart glasses value proposition. However, Meta could adopt this approach for images and video captured when you ask Meta AI to comment on what you're seeing.

Llamos asked if AI could shoulder some of the burden. “Here's an interesting solution: can AI tell the difference between an innocent interaction and one with ill intent?” Ramos said. “I don't know, and right now I don't think it can. That would require much training and learning, and the possibility for something to go wrong, like a false positive or not flagging an incident, is high.”

Brands are already thinking about how to address privacy issues, which is more about disclosure than anything else,” Milanesi said. “I say this because it is the people around you that need to be aware, not the person who is recording.”

What comes next?

The problem with smart glasses is as much behavioral as it is technical, argues Milanesi. “People just blaming the technology vs. the person are just thinking about poor use cases,” she said. “That is a human issue vs. a tech issue. What the brands must do is make sure the bad humans cannot do these things.”

It’s a point that Llamos agrees with. “Meta has repeatedly pointed to its suggestion to users to be more responsible in their use habits, but those have not proven effective,” he said.

Ultimately, Meta needs to prove it can be trusted — a tall order given its past indiscretions, but not impossible, especially if it can leverage the reputation of its other brands.

“People do trust some of Meta's properties — WhatsApp has had end-to-end encryption integrated by default for a decade,” Greengart said. “The combination of bad behavior by users and Meta's own lapses creates reputational debt that the company needs to tackle head-on.”

More from Tom's Guide

• Snap CEO Evan Spiegel says smart glasses disguised as normal eyewear are 'creepy' — why Specs look bold on purpose
• 'Pervert glasses' antidote? Meta reportedly launching camera-free smart glasses next week
• 'I find it very uncomfortable talking to people with a camera and a light': OpenAI's Sam Altman weighs in on smart glasses and privacy

Anthropic CEO Dario Amodei says frontier AI development needs to be paced and OpenAI CEO Sam Altman agrees. Elon Musk has backed the idea, as has Google DeepMind CEO Demis Hassabis.

And those warnings are pretty severe. Amodei wrote that AI is advancing "drastically faster" and pointed specifically to models' growing ability to help build the next generation of AI. His proposal includes giving independent evaluators deeper access to frontier labs, coordination between AI companies and international agreements around AI safety.

Then Michael Burry entered the conversation. The investor made famous by "The Big Short" has a very different explanation for the AI industry's sudden desire to hit the brakes. He argues that the warnings aren't simply about keeping humanity safe. He thinks they're also good for business.

Burry thinks the AI apocalypse is good marketing

Burry called the push to slow AI development "self-serving," arguing that OpenAI and Anthropic have strong financial reasons to portray their technology as extraordinarily powerful while also limiting how quickly competitors can catch them.

In the AI race, it's easy to overlook that big tech like OpenAI and Anthropic are so much more than research labs, they're companies competing in one of the fastest-moving technology markets we've ever seen; and the compeition is only escalating globally.

Burry argues that advances from Chinese and open-source AI developers are eroding some of the advantage enjoyed by the biggest American AI companies. Slowing frontier development, in his view, could therefore protect established players while making it harder for challengers to close the gap.

But there's another part of his argument that caught my attention. Because if you tell the world that the technology you've created is powerful enough to potentially threaten humanity, you're also telling the world that you've created an incredibly powerful technology. That's a pretty compelling message for a company eventually hoping to convince investors of its value.

The IPO timing makes this particularly interesting

Burry's argument would be easy to dismiss as another AI bubble warning if not for what's happening at OpenAI right now. Altman recently said OpenAI won't go public in 2026, as reported by Reuters, that the current focus on safety makes this an "ill-advised moment" for an IPO.

"We got a lot of stuff to do," Altman said, adding that OpenAI can deal with the current safety and alignment questions more easily while remaining a private company.

Burry sees something else in that decision by arguing that slowing development and delaying an IPO gives companies more time while simultaneously strengthening the narrative that their technology is becoming so powerful that extraordinary precautions are required.

That's an allegation rather than an established explanation for OpenAI's or Anthropic's behavior, and neither company's warnings about AI risk began this week. Anthropic in particular has spent years talking publicly about catastrophic AI risks and pushing for safety regulation.

But Burry isn't the only person questioning whether the proposed slowdown could benefit the companies calling for it.

Critics from across the political spectrum have raised concerns that an industry-led agreement could protect incumbent AI companies from competition or allow them to shape the rules governing their own technology. Former Federal Trade Commission official Alvaro Bedoya, for example, has argued that antitrust law shouldn't be used to let large AI companies coordinate in ways that keep cheaper competitors out of the market.

That doesn't mean the safety concerns aren't real

There are reasons AI researchers are legitamentally worried. I mean, this current debate didn't appear out of nowhere. It intensified after Anthropic researcher Jacob Coxon resigned and warned that leading AI companies were "gambling with our lives." Amodei subsequently pointed to recent security incidents and the rapidly improving ability of AI systems to contribute to AI development itself as reasons the industry needs to become more cautious.

There have also been concrete examples of models demonstrating capable behavior in cybersecurity and other potentially dangerous areas. Yet, Altman has said that pacing development shouldn't mean stopping it.

And this is where Burry's argument becomes more complicated.

There's an uncomfortable contradiction here

AI companies could genuinely believe their technology poses serious risks while also benefiting financially from telling the world how powerful that technology has become. Those things aren't mutually exclusive.

For consumers, the strange part is the position we're being asked to accept. The companies developing AI want us to trust them enough to put their models on our phones, computers and eventually inside agents capable of performing tasks on our behalf.

At the same time, some of those companies are telling us their models could eventually become powerful enough that the entire industry needs to deliberately slow down.

The debate isn't even settled inside Silicon Valley.

While Altman, Amodei, Musk and Hassabis have expressed support for pacing AI development, Meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang have pushed back against a coordinated slowdown. There are also enormous practical problems with getting competing companies — let alone competing countries — to slow down at the same time.

The U.S. government continues to view AI leadership partly through the lens of competition with China, while AI companies face enormous pressure from investors and customers to release better models. Even executives who agree that slowing down is necessary haven't yet solved the problem of how to do it without allowing somebody else to race ahead.

So should you believe the AI apocalypse warnings?

There isn't currently a reliable way to put a date on an AI catastrophe, and claims that AI will wipe out humanity within a particular number of years remain predictions rather than established facts. At the same time, dismissing every warning as marketing would go too far in the other direction.

What Burry's argument does is add another question worth asking whenever an AI company tells us how dangerous its technology could become. What does the company gain from us believing it?

When a company tells the world that its technology could become powerful enough to change civilization — or even threaten it — it's hard to know if it's a warning or a heck of a sales pitch.

More from Tom's Guide

• AI is turning ordinary photos of kids into sexual abuse imagery — here's how to protect your family
• OpenAI paid contractors to read ChatGPT conversations — here's how to protect yourself
• I stopped reading the fine print myself after discovering this overlooked ChatGPT trick

On Tuesday (Sept. 15), Meta officially launched Meta One, a new subscription service that bundles premium features across Instagram, Facebook and WhatsApp while giving subscribers higher limits for some of the company's most compute-intensive AI tools.

The company says Meta One plans are available globally, although pricing, features and availability can vary depending on your region, app and account.

You definitely don't need to subscribe to keep scrolling Instagram, messaging friends on WhatsApp or using Meta AI normally. Meta says the core experiences across its apps and Meta AI will remain free.

The catch is that some of the more demanding AI features now come with usage limits. Hit one, and you could be asked to pay if you want to keep going.

So what exactly are you paying for?

Interestingly, Meta One isn't a single subscription with a single price.

In the U.S., Meta offers separate Plus plans for its individual apps: Instagram Plus costs $3.99 a month, Facebook Plus costs $3.99 and WhatsApp Plus costs $2.99. Each adds its own collection of extras.

Instagram Plus, for example, includes features such as custom fonts for DMs and Stories, notifications when certain people view your Story and DM previews. WhatsApp Plus includes exclusive themes and icons, special-effect stickers and the ability to pin more chats.

Meta's AI tools are Meta One Core and Premium

The plans for anyone using Meta's AI tools are Meta One Core and Premium. Core costs $7.99 a month and includes the features from all three individual subscriptions. More importantly, it increases how often you can use features including Instagram Restyle and voice effects, as well as how many images and videos you can generate with Meta AI.

Premium jumps to $19.99 a month and includes everything in Core with even higher AI usage limits.

In other words, Meta isn't necessarily charging you to use its AI. It's starting to charge its heaviest users to use more of it.

How soon will you hit a paywall?

Despite having the subscription and using the AI tools pretty regularly for experimenting and reviewing, I have not hit a paywall yet, but Business Insider's Sydney Bradley encountered the limit while experimenting with Instagram's AI photo-editing effects. After repeatedly using the tools, Instagram presented an option to upgrade to Meta One Core for additional AI usage rather than continue using the feature freely.

If you occasionally use Meta AI to answer a question or try an Instagram AI effect, Meta says everyday Meta AI use remains free. But start generating or editing enough images and videos and you may find there's now a ceiling on how much you can do without paying.

Meta hasn't published one simple universal number telling users exactly how many AI generations or edits they get for free. The company says plans, benefits and availability can vary by account and region.

That could make the change particularly noticeable for people who don't realize they're approaching a limit until Instagram asks them to subscribe.

Meta One is about more than Instagram

Meta One also gives us a pretty good idea of where Meta is heading.

The company says more than 50 features have already launched as part of the broader Meta One rollout, which has accumulated 15 million subscriptions and trials so far. The service will eventually expand beyond Instagram, Facebook, WhatsApp and Meta AI to include Edits and Meta's AI glasses.

There are separate Meta One subscriptions aimed at creators and businesses, too, with features including analytics, publishing tools, verification and greater access to Meta Business Agent. Those plans start at $14.99 a month and climb dramatically from there, with Meta's highest business tier starting at $499 a month.

For most people, though, Core and Premium are the plans that make the most sense.

Bottom line: Do you need Meta One?

Probably not — at least not yet. I have it because, well, I tinker around with AI a lot (it's literally my job). If you primarily use Instagram to scroll through Reels, Facebook to keep up with friends and WhatsApp to send messages, nothing fundamental is disappearing behind a subscription.

To be clear, Meta AI isn't becoming a paid-only service. What Meta is doing instead is drawing a (somewhat fuzzy) line between using AI and using a lot of AI. It's no different than what Google or OpenAI offer because image and video generation demand requires considerably more computing power than ordinary social media features.

Meta One gives the company a way to let everyone try those tools while charging the people who use them most. Whether consumers think Instagram's AI effects are worth another $7.99 a month is another question entirely. Meta clearly thinks some of us will.

During early testing, the company says more than half of subscribers used both AI and expression features, with Instagram's Restyle and voice effects among the biggest reasons people signed up. And with 15 million subscriptions and trials already under its belt, Meta may have found another subscription we're willing to pay for on apps we've spent years using for free.

More from Tom's Guide

• AI is turning ordinary photos of kids into sexual abuse imagery — here's how to protect your family
• OpenAI paid contractors to read ChatGPT conversations — here's how to protect yourself
• I stopped reading the fine print myself after discovering this overlooked ChatGPT trick

You can even give it a bigger goal and let it figure out the steps required to get there, which makes Muse one of the clearest examples of where consumer AI appears to be headed.

But there’s an obvious tradeoff: For Muse to become genuinely useful, you have to give it access to a remarkable amount of your digital life.

Muse isn't just another chatbot

Meta describes Muse as a “personal AI agent,” which is similar to what we've seen from Gemini and ChatGPT recently. Like the AI agents from Meta competitor's Muse is designed to do more.

According to Meta, the agent can open websites, fill out forms and negotiate on your behalf. It can book travel, send emails and make purchases, checking back with you when it needs approval for certain actions.

It can also keep working after you close the app, which opens the door for much longer-running tasks. This could include tasks such as trying to lower a bill, adapting a training plan when your schedule changes or even selling a car, according to Meta's examples in the company press release.

Muse can also remember information you've previously shared and use it later. Meta says it could notice a recipe Reel you saved on Instagram, turn it into a grocery list and remember your friends' dietary restrictions when helping organize a dinner party.

In other words, the goal isn't necessarily for you to keep prompting Muse. It's for Muse to learn enough about your life that it can start anticipating what you might need.

It can spend money for you

One of the biggest differences between an AI assistant and an AI agent is what happens when money gets involved. Muse can make purchases using Stripe's Link payment system. Instead of giving the agent your actual card number, Link generates a one-time-use card for the transaction.

Meta says Muse is also the first AI agent covered by Link's purchase protections, which can include things such as no-fee returns and protection for eligible lost or damaged items.

Shop Pay support is expected to arrive later, while Meta says integration with 1Password is also coming so Muse can use credentials stored in a user's password manager.

That could make tasks such as shopping considerably more convenient. It also raises the stakes considerably if an agent misunderstands what you asked it to do.

Meta built a second AI to watch the first one

Meta clearly knows that giving an autonomous AI access to email, accounts and payments creates an entirely different security problem. Muse therefore runs inside what Meta calls a Secure VM — essentially its own isolated computer in the cloud.

There's also another AI agent running alongside it called Sentinel, separated from Muse at the system level and reviews what the agent is trying to do before its actions reach the internet. Meta says Muse also asks for permission before sensitive actions such as sending an email or making a purchase.

Users can decide which services Muse connects to and what permissions it receives. Giving Muse access to email, for example, doesn't necessarily mean allowing it to send messages.

There's also an audit trail showing what the agent has done and what it plans to do. Meta says Muse doesn't see passwords or payment information directly. Credentials are kept in secure storage and can be used by the agent without revealing the underlying information to it.

Later this year, Meta plans to introduce what it calls Muse Confidential VM, which will encrypt the entire virtual machine with a key held by the user. Meta says that would prevent even the company itself from accessing the data and conversations stored inside.

Safeguards and trust

All of those safeguards matter because Muse potentially has access to some of the most sensitive parts of someone's digital life. And Meta's own testing shows why autonomous agents remain difficult to secure.

Reuters reports that Meta originally planned to release Muse earlier this year but delayed the launch while it worked on security.

Internal testing reportedly produced mixed results. In one particularly concerning example, an employee said the agent managed to get around safeguards and expose personal iCloud photos after being asked to identify toys visible in pictures from a child's birthday party.

Other employees reported reliability problems, including an agent that stopped refreshing a webpage while monitoring hard-to-find tickets and sometimes silently ignored errors.

Meta told Reuters that it had done enough additional security work to reach the threshold required to release Muse, although the company acknowledged that mistakes remain possible.

Final thoughts

We've now entered a post-chatbot era. There are already plenty of AI agents capable of using computers, browsers and other tools. What's potentially significant about Muse is who Meta wants using it.

Muse is rolling out in the U.S. on iOS and Android and through the web, and you can also talk to it directly through WhatsApp. Meta says support for its AI glasses is coming soon. There's a free tier, with paid plans available for people who need heavier usage.

More from Tom's Guide

• AI is turning ordinary photos of kids into sexual abuse imagery — here's how to protect your family
• I asked ChatGPT, Gemini, Claude and Perplexity to rank the 10 jobs most likely to be replaced by AI — here's what they all agreed on
• Gemini just got a 'floating bubble' multitasking upgrade that's a game changer on Android phones — here's how to use it

332 AI-generated child sexual abuse ads slipped through

A new investigation from the Tech Transparency Project (TTP) found hundreds of ads containing AI-generated child sexual abuse material (CSAM) running across Facebook and Instagram. Researchers identified four real children whose ordinary photos had been manipulated, according to WIRED, including two teenage influencers, a preteen whose image appeared on a stock-photo site and a young member of a European royal family.

In total, TTP found 332 ads containing AI-generated child sexual abuse material, or CSAM. Most depicted a child whose photograph had been manipulated using AI to make it appear they were performing a sexual act.

The findings are an alarming reminder of how generative AI has changed the risks around sharing photos of children online. Creating convincing fake imagery no longer requires an explicit photograph in the first place. An ordinary picture can provide the raw material.

How ordinary photos become AI-generated abuse

Generative AI image tools can take an existing photograph and modify parts of it while preserving recognizable features of the person in the original.

That's what researchers say happened to several of the children identified in the Meta ads.

According to WIRED, researchers were able to trace images of four minors back to their original sources. Two were teenage influencers with public social media accounts. Another was a preteen whose photograph was available through a stock-photo website. The fourth was a young member of a European royal family.

None of those original photographs needed to contain sexual material for them to be misused.

"AI apps are making it far too easy to create realistic criminal imagery of child sexual abuse, creating the new risk of everyday photos of children being taken and transformed into sexual content," Emma Hardy, director of communications at the Internet Watch Foundation, told WIRED.

Once an ordinary image is publicly accessible, someone can potentially download it and feed it into an AI tool designed to alter photographs.

Meta's AI didn't catch hundreds of the ads

The scale of the latest investigation also shows how difficult the resulting content can be to stop.

In August, TTP researchers initially uncovered 53 ads containing child sexual abuse material across Meta's platforms. Meta subsequently said many of those ads had appeared before the company introduced new AI technology intended to improve its ability to detect violating ads when they were uploaded.

However, researchers kept finding them. In fact, TTP says it ultimately identified 332 ads containing AI-generated CSAM, with hundreds appearing after Meta had announced its improved detection technology.

That is particularly significant because these weren't simply Facebook or Instagram posts that escaped moderation. They were paid advertisements.

Meta says ads are reviewed before they're allowed to run and that automated systems play a major role in that process.

Meta spokesperson Tracy Clayton told WIRED that the company does not tolerate nudify apps or child exploitation, whether the imagery involves real children or is generated by AI. Meta said bad actors deliberately attempt to evade its detection systems and that some of the videos were harder to detect because images of children appeared only briefly or were blurred.

The company said many of the ads researchers identified had already been removed, most received fewer than 200 impressions and it received less than $5,000 in payments from the ads. But researchers documented instances in which exposure increased after an ad was reported.

TTP says one ad involving a real teenager had reached four European accounts when it was reported to Meta. Several days later, when Meta removed it, the ad had reached more than 330.

The investigation shows the tools are mainstream

Researchers found ads directing people toward nearly 50 apps available through Apple's App Store and Google's Play Store. Around 300 of the roughly 350 ads examined in the broader investigation promoted AI image or video generators, according to WIRED.

Apple told WIRED that some developers initially submitted compliant versions of their apps before subsequently adding prohibited nudification capabilities. Apple said it removed 17 of 20 apps identified by researchers and took action against the remaining three, as well as seven additional apps WIRED identified.

Google said it removed all apps included in the report and suspended hundreds of other apps for violating its policies.

Still, these finding reveal a problem that extends beyond any single platform because AI can be used to create the imagery. An app store can distribute the tool used to make it. An advertising platform can then distribute ads promoting it.

Each layer may have its own safety systems, but abusive material only needs to slip through those systems once to reach users.

What to do if you find an AI-generated image of your child

If you discover sexualized AI imagery involving your child, don't repost or widely share the image in an attempt to warn others.

Instead, preserve information that can help identify where the material appeared, such as the account name, page address and other surrounding details, and report the content directly to the platform.

In the U.S., suspected child sexual exploitation can also be reported to the National Center for Missing & Exploited Children's CyberTipline.

Parents can also use NCMEC's Take It Down service, which is designed to help stop the online sharing of nude, partially nude or sexually explicit images or videos involving someone who was under 18. The system creates a digital fingerprint, or hash, from an image without requiring the image itself to be uploaded.

Since AI-generated imagery creates newer and more complicated scenarios, parents who aren't sure whether a manipulated image qualifies should still report what they've found rather than trying to determine that themselves.

The takeaway

Parents have been warned about "sharenting" and children's digital footprints for years. Sharenting is a term that combines "sharing" and "parenting" and is the practice of parents sharing images, videos and personal details about their children on social media.

Although parents can't entirely eliminate the risk of AI abuse on the images or videos they post, but reducing the number of publicly accessible photos of your children and being more selective about what you share, can reduce the amount of material available to people who might misuse it.

More from Tom's Guide

• ChatGPT-6 Astra is here — this one prompt shows what it can really do
• I asked ChatGPT, Gemini, Claude and Perplexity to rank the 10 jobs most likely to be replaced by AI — here's what they all agreed on
• Claude Fable 5.1 is here — this one prompt shows what it can really do

The small white LED on the front of Meta’s glasses flashes whenever the camera is taking a photo or recording video. It is supposed to give anyone nearby a visible warning that they may be on camera.

However, users have found ways to obscure or physically modify that light. Meta’s latest update is designed to stop the camera from recording if someone begins filming and then covers the LED.

This closes a potential loophole in the tamper-detection protections Meta announced in July. Beginning with the second generation of its glasses, Meta says the camera is disabled when the device detects that the capture LED has been blocked. The company later expanded that protection to detect attempts to physically alter or destroy the light.

Now the glasses are also expected to recognize when the LED becomes covered during an active recording session and shut the camera down.

Meta is going after tampering services

The technical update is part of a much broader effort to discourage people from using Meta’s glasses to record others without their knowledge.

Some companies had offered services that modified or concealed the capture light. Meta says businesses offering those services have begun shutting down because its glasses now deactivate the camera when tampering is detected.

The company has also removed posts offering instructions or services for disabling the LED.

Meta says it has taken down thousands of posts connected to harassment, pick-up artist videos and prank content, while disabling accounts involved in that behavior. It has also blocked searches and hashtags associated with violating content, including “rizz,” “pick up artist” and “cold approach,” in an attempt to make those videos harder to find and share.

Meta wants everyone to recognize the light

Wearing a camera on your face makes it far easier to film someone without holding up a phone, particularly if that person does not recognize what the flashing white light means.

Alongside the update, Meta is launching an awareness campaign explaining what the capture LED is and what it means when it starts flashing.

The campaign will include online videos and images as well as large billboards in communities where sales of the glasses have been especially strong. One is already running in Los Angeles.

That education may be just as important as the technical safeguard. The LED can only provide a useful warning if the people around the wearer understand why it is blinking.

Meta’s glasses are becoming far more common, but they are still unfamiliar enough that someone across a restaurant, playground or commuter train may not realize they are looking at a wearable camera.

The latest update will not settle every privacy concern surrounding smart glasses. Someone can still record people who have not explicitly agreed to appear in a video, just as they can with a smartphone. But making the light harder to defeat, while helping people recognize it, should make covert recording more difficult.

More from Tom's Guide

• Perplexity’s new local-first AI runs on your PC — and asks before using the cloud
• ChatGPT can learn your repetitive tasks by watching you — here are 7 prompts to try
• The best way to use ChatGPT for writing isn't writing — 7 prompts I use to make it my creative partner

That initial dream of mine quickly turned into something I know I wasn’t built for once I was left utterly confused after reading up on the topics of C++ and C# for game engines and system logic. To be completely honest and blunt, I just knew I wasn’t smart enough to grasp everything that goes into creating the types of games me and my friends wanted to play.

But as I sit here and type out this article, I feel like I finally get to achieve that lofty goal of mine in the strangest way possible — through a vibe coding app that’s accessible from the comfort of my iPhone. Meta recently launched a free app called Pocket for iOS and Android devices that lets you get as creative as you’d like as a mobile game maker and prompt your concept into something that’s actually playable.

I figured it was time to show the world my creative genius as a game designer by vibe coding (which is simply describing what you want to an AI instead of manually writing out code) my own playable experiences. And honestly, this little experiment ended up being more than I thought it would.

Slashing unicorns, shooting bubbles and bouncing on clouds with trampolines

During my first go-round with Pocket, I was treated to the Pocket app’s default game, which told me to tilt my phone to move a gumball machine full of sweet, multicolored contraptions. It was a decent little interactive experience, but I yearned for something a bit more complex and satisfying to play.

For my first game, I had the inkling to get into some medieval fantasy action suitable for younger players. Once I put in my prompt, the app went to work for a few seconds as it turned my words into a playable game. Imagine my amazement when the finished product came out as something that would have kept me busy during my days as a toddler who had no siblings to annoy.

All I had to drag my pint-sized warrior across the screen warrior across the screen and tap it whenever I wanted to swing my blade to vanquish some no-good unicorns. After spending far too much time cutting down evil creatures falling out of the sky, I started my next game project.

This time, I felt like honoring a classic match-three puzzler that got far too many quarters out of me back in the day. Bust-A-Move (better known as Puzzle Bobble in Japan) is simple, cute and straight to the point — attach three bubbles of the same color to make them pop so you can clear them all out. This type of game has gotten a million and one copycats across consoles and mobile gaming storefronts.

My prompt explained that game’s concept quite clearly and, thankfully, the Pocket app knew exactly what I was trying to make. 15 seconds later and BOOM…my own match-three bubble puzzle game was born. Tapping to shoot my bubble into other ones became my dosage of fun for about 45 minutes before I had another lightbulb moment for my next game creation.

For my third and final vibe coding exercise, I wanted to play something that’s all about staying in the air and not falling to my doom.

With visions of Jetpack Joyride circling my brain, I entered my prompt and ended up producing a wholesome game I’d let my younger cousins play all day and night. All I had to do to keep my miniature avatar alive was tilt/drag the screen to move them from left to right and have them bounce on trampolines encased within clouds. The blue and white ones gave me a decently sized jump, while the grey and white ones disappeared after bouncing on them once and the pink and yellow ones threw me even higher into the sky.

This game ended up being my most complex since it featured some challenging elements that kept it from getting stale after a minute or two.

Once I was done making my games, I took a trip to my “Playlists” tab to see if I could find other people’s games. All I had to do was scroll down and find the “Discover” section full of curated picks to find even more interesting ideas. Plus, every time I opened the Pocket app after a few hours, a whole new game would pop up on my home screen (shout out to “josh” for introducing me to his highly entertaining “pocket space copter” game).

Final thoughts

Meta’s Pocket app is such a fun way to familiarize yourself with vibe coding and turn your video game concepts into something that’s actually tangible. To keep my creative juices flowing, I went to ChatGPT to see if it could generate some game-making prompts (one of them gave me the perfect Flappy Bird copycat!), so feel free to use these for yourself and add your own details to make them all your own:

• Flappy Bird-style: Create a simple one-button arcade game where I control a tiny UFO flying through a constantly changing alien city. Tap to rise and release to fall. Add increasingly difficult obstacles, a score counter, high-score tracking, and funny sound effects.
• Angry office printer: Create a game where I'm an office worker trying to defeat an evil printer. The printer randomly jams, spits out paper, launches ink cartridges, and summons error messages. Give me different attacks and make the boss progressively more ridiculous.
• Don't press the button: Create a game centered around a giant red button labeled "DO NOT PRESS." Every time I press it, something different and increasingly absurd happens. Track how many times I've pressed it and create at least 20 different outcomes.
• Choose-your-own-adventure RPG: Create a visual RPG where every decision I make changes the story. Give me health, inventory, money, and reputation stats. Include at least 10 different endings and make some endings only accessible through specific combinations of choices.
• Let Pocket make something on its own: Build me a simple playable arcade game based on a concept you invent. Keep the controls extremely simple and make the first version playable within 30 seconds. After building it, add three mechanics that make the game more addictive without making it significantly more complicated.

More from Tom’s Guide

• Perplexity’s new local-first AI runs on your PC — and asks before using the cloud
• The best way to use ChatGPT for writing isn't writing — 7 prompts I use to make it my creative partner
• I use these 5 Gemini prompts before any vacation to research my destination — and they’ve saved me time and stress every time

The company’s smart glasses have gotten a ton of ire as people have pointed to privacy concerns over users secretly filming random passersby and whoever else they please (there’s even a report about teenage boys using them to pester girls at middle and high schools).

Meta then found itself in the middle of a lawsuit that saw 29 states, co-led by California Attorney General Rob Bonta and the AGs from Colorado, New Jersey and Kentucky, allege that its social media platforms are addictive to children, mislead consumers about those young users’ safety on said platforms and inappropriately gather children's personal data.

That lawsuit has now met its end, as Meta has just agreed to pay a settlement of $17.1 billion — according to a statement made by California AG Rob Bonta. The state of California could receive between $1.5 billion and $2.1 billion if the court authorizes the settlement. As for New York, its attorney general announced that the state will get up to $1.15 billion and likely put it toward kids' education and social services meant to assist youngsters struggling with their social media addiction. Head to this link to see how much each state will receive from this monumental Meta settlement.

With this settlement in place comes a host of changes Meta has been ordered to make to Facebook and Instagram. Here’s what these changes could mean for parents with children who frequently use both platforms and how they could result in a positive outcome for them.

Less addictive and harmful social media spaces for kids of all ages

There’s a wealth of changes Meta must now make according to the measures detailed in the settlement, such as:

• A default daily time limit of two hours for users under 18 years of age that can only be lifted by a parent
• A default nighttime block between midnight and 06:00 for users under 18 years of age that can only be lifted by a parent
• Default blocks on notifications to users under 18 years of age from 22:00 to 07:00 and during the school day
• An enhanced mechanism for teenagers to report potentially harmful content and a requirement that Meta respond to 90% of those reports within six hours
• A ban on displaying numbers of likes or reactions to users under 18 years of age
• A ban on cosmetic procedure image filters for users under 18 years of age
• An option for users under 18 years of age to have a non-personalised feed
• A commitment to enhance parent supervision tools

That settlement also notes that Meta has agreed to the following terms and conditions:

• Institute robust age assurance measures to detect users under 18, as well as measures to identify and remove users under 13 years of age
• Use an independent auditor with expansive access to information and resources
• Be subject to an injunction prohibiting it from making "further false, misleading, or deceptive statements around its safety features"

In a statement that looks to reassure everyone that kids will be more protected across their social media platforms, Meta had this to say: “Ensuring teens have a safe and productive experience on our platforms is an absolute imperative for Meta. We want to get this right for parents and teens.”

With all these newly instituted measures in place, Facebook and Instagram could become far less of an addictive haven for children who want to do nothing but endless scrolling and discover content that could be extremely harmful to their mental well-being. Strict usage time limits, parents getting more control over the type of content their kids see, fewer distracting notifications popping up on kids’ screens and young users having less exposure to detrimental content could go a long way toward making Meta’s platforms more approachable for younger users.

While instituting a two-hour limit for kids using Facebook and Instagram is good, kids may still use that time to expose themselves to harmful content. With parents getting more control over their kids’ social media accounts, here’s hoping their stronger involvement will make children less susceptible to coming across said content.

Bottom line

Meta being forced to take responsibility for the harm its social media platforms have done to children is major. While this development doesn’t automatically make Facebook and Instagram 100% safe for kids from this point forward, it’s a good first step toward companies like Meta recognizing the ill effects their products have had on children.

More from Tom’s Guide

• Perplexity’s new local-first AI runs on your PC — and asks before using the cloud
• ChatGPT can learn your repetitive tasks by watching you — here are 7 prompts to try
• AI could transform healthcare for the better — but this doctor says there’s one thing it should never do

A new experiment by Compare the Market tested exactly how much information can actually be hidden when you set your Instagram to private. The results are sobering.

A licensed investigator was tasked with locating the home addresses of nine volunteers using only publicly available information.

Despite four of them having private Instagram profiles, the investigator identified the addresses of seven out of nine. The fastest was located in just 15 minutes, with the first lead found in 50 seconds.

What the investigation revealed

Colin Crowden, Intelligence Director at Global Protect and an Open-Source Intelligence expert, conducted the investigation. What he discovered challenges the assumption that private accounts provide adequate protection.

Beyond identifying addresses, the investigation uncovered four complete property plans revealing internal home layouts. In one particularly concerning finding, Crowden established that a volunteer was away on holiday, confirming their property was unoccupied.

The speed at which information was gathered highlights a critical vulnerability. Information shared by friends and family, location tags from previous posts, property details visible in photos, and data from other online sources all contribute to a broader picture.

Crowden explains the issue clearly: "Information that appears harmless in isolation can become much more revealing when viewed alongside other publicly available information. This isn't about any one post creating a security risk; it's about how different pieces of information can build up over time."

Why your private account isn't enough

The experiment demonstrates that privacy settings alone offer incomplete protection. Even if your Instagram is private, tagged photos from friends, public check-ins, or incidental details in backgrounds can expose your location and home information.

The investigation reveals that the combination of data sources, not any single post, creates the vulnerability. Compare the Market's home insurance expert Helen Phipps worked with Crowden to develop concrete recommendations for reducing your digital footprint.

Practical steps to protect yourself

Wait until you're home to post

If possible, wait until you're home before posting holiday photos rather than sharing in real time. This prevents anyone from knowing your property is unoccupied. If you do post while away, avoid revealing exactly where you are or how long you'll be away for.

Review your privacy settings before you travel

Check who can see your posts, stories, and tagged content. Consider setting your accounts to private and reviewing your followers so that holiday updates are only visible to people you trust. This step takes minutes but significantly limits exposure.

Coordinate with friends and family

Your friends and family can inadvertently expose you. Ask them not to share real-time updates if you're traveling together or post photos of your property if they're housesitting.

People often don't realize they're compromising your security by tagging you in posts or sharing location details.

Turn off automatic location tags

Disable geotagging in your social media settings entirely. On Instagram, this requires revoking access directly through your phone’s system settings under Privacy & Security (iOS) or Apps (Android) rather than within the app itself.

Location sharing can unintentionally reveal your whereabouts even when you're not posting in real time. This is one of the easiest settings to overlook but also one of the most revealing.

Audit your digital footprint

Search for yourself online periodically to see what personal details are publicly accessible, and remove anything that compromises your privacy. Pay special attention to old social media posts containing house numbers, street signs, car license plates, or identifiable landmarks.

If you want to take action right now, I have a step-by-step walkthrough covering 5 ways to start removing your personal information from the internet. You should also check out how to blur your home in Google Maps Street View.

Combine digital caution with home security

Beyond digital precautions, pair online caution with classic home security measures. Set light timers to create the appearance someone's home, test all locks, arm your alarm system, and ask a neighbor to collect post and manage bins so the house looks lived-in.

Ensure your home insurance policy covers the duration of your trip and review any conditions that apply while your property is unoccupied.

This investigation raises important questions about privacy in the digital age. Have you changed your social media habits? What's your biggest concern about your digital footprint? Share your thoughts in the comments below.

For a complete breakdown of Compare the Market's methodology and findings, read the full Online Safety Study here.

More from Tom's Guide

• Meta's new AI tool uses your Instagram photos by default — here’s how to opt out
• Inside the dark market stealing real identities online
• Instagram's 'instants' feature is driving users crazy. Here's how to disable it

As reported by The New York Times, acting ICE Director David Venturella told agency employees in a memo to stop wearing Meta's camera-equipped glasses because they count as personal body-worn cameras and cannot be worn in federal workplaces.

"The use of Meta Glasses or similar devices could unintentionally capture, record, or transmit sensitive information, potentially compromising privacy and legal protections," Venturella reportedly wrote.

The ban apparently applies to all ICE employees, not just those agents conducting raids or interacting with the public. It covers all personal devices capable of recording audio or video.

Not the first time

A Department of Homeland Security spokesperson told the Times that this isn't a new policy. "Personally owned body-worn cameras and unauthorized recording are prohibited, as they always have been."

Concern over ICE agents wearing Meta glasses isn't new either. Last year, 404 Media reported on Border Patrol agents who were photographed wearing Meta glasses clearly in recording mode.

In March, The Independent investigated multiple sightings of DHS agents wearing Meta glasses in at least six different states. In some of those instances, agents appearred to be using hte glasses to photograph or record members of the public.

Part of the concern, at least on the government's part, is that personal devices mean the government can lose control of the footage. The videos could be passed on to a personal phone or filter through Meta's platform with no oversight of who can access it or what information is actually in the recording.

And that's all before we get into reports that DHS wants to design its own smart glasses. As reported by Gizmodo in April, the agency wants the glasses specifically for surveilling "illegal aliens." This version would use existing smart glasses combined with biometric databases and facial recognition to "identify people in real time."

Meta's privacy problem

Even outside of government surveillance, privacy concerns are an ongoing problem for Meta and its smart glasses. Recording can be quickly started, and the LED light indicator can easily be circumvented. Plus, there's the report from July that Meta is building "super sensing" glasses that would continuously record your life.

Meta is no stranger to pushing through privacy concerns. The company was sued over privacy claims in March over claims that workers in Kenya were watching intimate footage captured by Meta smart glasses.

In 2025 alone, the company filed a suspicious patent that would allow AI facial recognition glasses to identify people and record automatically, was caught tracking Android users without consent, accused of accessing camera rolls without permission for AI training, and was accused of displaying revealing personal info in the Meta AI discover feed.

Meanwhile, Meta currently faces a $1.4 trillion lawsuit over its claimed failure to protect children on its social platforms like Facebook and Instagram.

More from Tom's Guide

• Meta files patent for AI facial recognition glasses that identify people around you and record automatically
• Meta's new AI tool uses your Instagram photos by default — here’s how to opt out
• Meta Connect 2026 kicks off in September — date, time and mystery new smart glasses teased

When a tech giant is put on trial, the public is usually conditioned to expect a lengthy battle that ends in a quiet settlement, written off as just the cost of doing business. But this is entirely different.

That’s because after the lawsuit was filed by 30 U.S. states in 2023, including California and New York, governments are finally taking action in saying the “attention economy” is a public hazard for children.

It’s not just an accusation of bad content policy, it is a direct assault on how Meta and social media as a whole works. Beyond that eye-watering figure, the states are demanding fundamental changes to how Instagram and Facebook operate — including killing the like button and infinite scroll.

Should Meta lose, this would be a massive tectonic shift. I’ve sifted through the entire case document to break down the accusations, the demands, the evidence and what this will actually mean for you.

The hit list (what is actually changing)

I’ll go into the specifics of the case itself in a little bit, but outside of that gigantic price tag, what would actually vanish from Meta’s apps if the company loses?

• Death of infinite scroll & autoplay: The states are demanding an end to the endless feed and autoplaying of content. These features are explicitly designed to maximize screen time, and losing them breaks the TikTok-style doomscroll in a fundamental way.
• The end of the like button: Another key dopamine hit that keeps teens hooked to their screens is the like count. The lawsuit is targeting this too — demanding an end to social validation metrics.
• No more algorithmic manipulation: The states want to see change to Meta’s “dopamine-manipulating recommendation algorithms.” This could mean a return to your standard chronological timelines by default.
• Notification lockdown: No more frequent push notifications designed to pull young people back into apps. You’ll see these pop up to stop young people from disengaging with a social platform.
• Killing stories & filters: The states want an end to disappearing or “ephemeral” posts like Instagram Stories. On top of that, they’re demanding the removal of image filters that alter a user’s physical appearance in pictures.

However, there is one broader, scarier implication for everyone, not just teenagers.

The death of the “Finsta”

The lawsuit isn’t just about tweaking algorithms; it’s demanding structural changes to how accounts are created. The states are demanding strict parental verification for teenagers and limits on multiple account creation.

For teens, this effectively kills the “Finsta” — a fake or secondary Instagram account used to hide from parents or public scrutiny), and to enforce this, platforms will need to know exactly who is holding the phone. That would spell the end of casual online anonymity.

The smoking gun — could Meta lose this case?

Obviously, I’m not going to claim myself to be some expert legal mind here. But I’ve been a tech journalist for over 15 years and witnessed a bunch of lawsuits, and some of the data points, internal admissions and quotes in this document seem pretty eye-opening.

The COPPA Scandals:

For over seven years (until December 2019), Instagram had no age gate at all. When Meta finally added one, the drop-down menu defaulted automatically to a date making the user exactly 13 years old — effectively showing kids the exact date to bypass the restriction.

Internal research cited in the filing revealed that 40% of children aged 9 to 12 use Instagram daily, and 45% use Facebook daily. This leads to the accusation that Meta viewed children aged 10 to 12 as a “valuable but untapped audience” that dedicated internal teams analyzed to target.

The founding confession

Sean Parker himself said that the thought process behind building Facebook and Instagram was all about “how do we consume as much of your time and conscious attention as possible?”

He called it a “social-validation feedback loop” that exploits “a vulnerability in human psychology,” which its creators all understood “consciously.”

“I don’t want to hear it” about underage users

This is what Head of Instagram Adam Mosseri apparently said when he was confronted with evidence that child star JoJo Siwa had been active on Instagram since age 8.

On top of that, Meta’s former director of site integrity Arturo Bejar testified that Zuckerberg knew about “harms that teenagers are experiencing in its product,” but are “choosing not to engage about it.”

Meta’s own internal numbers

The company ran a “Teen Mental Health Deep Dive” private survey that found:

• 1 in 5 teens felt worse about themselves after using Instagram
• 24% reported feeling “not good enough because of the app”
• 67% felt forced to “create a perfect image online”
• Over half admitted to struggling with severe FOMO (Fear of Missing Out)
• 14% of young users admitted to having thoughts of wanting to “hurt [or] kill themselves”

Cynical product design and internal teams

The lawsuit pulls back the curtain on how Meta internally viewed user engagement — with the company reportedly operating an engineering group called the “Rabbithole team.” Their mission was clear: optimizing “preference amplification,” which is fancy wording for figuring out how to push users deeper down algorithmic rabbit holes.

Even when Meta rolled out “time management” tools like the Daily Limit to appease public critics, the lawsuit alleges they were designed to fail. Originally, the daily limit was 10 minutes, but once public scrutiny died down after the 2021 whistleblower leaks, Meta quietly raised that minimum to 30 minutes.

But perhaps most damning of all is that as early as 2017, Facebook’s own former VP for user growth publicly admitted he had banned his own children from using Facebook. Sometime after that, Meta simply disbanded its internal “Responsible Innovation” team.

Outlook: a corporate death penalty?

The federal trial officially kicks off today in Oakland, California, with California, Colorado, Kentucky and New Jersey leading the charge.

But this trial isn’t just about the money. Because this is a consumer protection lawsuit rather than a personal injury case, the states are asking the judge to order sweeping changes to the platforms.

If Meta loses, we aren’t just looking at a massive fine. We are looking at court-mandated age restrictions, the deletion of algorithmic feeds for minors, and the permanent removal of the infinite scroll.

Win or lose, the curtain has been pulled back. The “attention economy” as we know it is officially on trial.

More from Tom's Guide

• Meta files patent for AI facial recognition glasses that identify people around you and record automatically
• A parent’s guide to the MacBook Neo: is Apple’s cheapest laptop a back-to-school trap?
• I connected Claude to my daily apps — here are the 5 that actually made it useful

The names Sam Altman (OpenAI), Satya Nadella (Microsoft) and Dario Amodei (Anthropic) are among those individuals who’ve shared their insights into the trajectories of continued AI development, their impact on society, the long-term health/scientific benefits of advanced AI, etc. While some have found their opinions on the growth of AI and predictions for how they’ll make us better fascinating, others have found them to be full of fantastical promises that will never come to fruition.

Another individual whose name joins the list of “tech bros who’ve written lengthy papers about AI” is Mark Zuckerberg, the CEO and founder of Meta Platforms (better known as the guy who made Facebook).

He just dropped a 6,500-word AI manifesto titled “The Future is for Everyone: The Path to a Positive AI Future.” I got real comfortable in my office chair, pushed my glasses up to my nose and locked in as I read it to completion to find the five biggest takeaways you need to know.

Personal superintelligence for everyone

I’ll break it down for you guys and gals reading this right now via five easily comprehensible bullet points:

• Everyone should have access to their own “personal superintelligence”: Zuckerberg doesn’t think AI should solely belong to governments, corporations or AI Labs— he feels it needs to belong to anyone and everyone. He wants to live in a world where the everyday person has a highly capable personal AI agent at the ready that comprehends their goals and works tirelessly to help them with tasks associated with their health, career, finances, relationships, and hobbies.
• Invention over automation: Zuckerberg pushes back against the notion that AI’s continued evolution will result in more human jobs being eliminated. He is more of a huge proponent of AI improving our lives by helping humanity discover new life-saving drugs, find new ways to improve businesses and produce new creative ideas.
• A single benevolent AI is not the right way forward: Zuckerberg isn’t in favor of the concept of a single small group perfecting and controlling one single aligned superintelligence. He believes the distribution of power, such as checks and balances between individuals, businesses and governments that rely on AI power, should be introduced to ensure more safety for AI usage. For example, if only one person had a superintelligent lawyer, they'd have an unfair advantage, but if everyone has one, justice becomes fairer and more achievable.
• Meta is in the process of building the best personal superintelligence: Zuckerberg outlines several ways in which his company Meta is developing the sort of AI that will benefit humanity in the long run. Meta is focused on developing a capable personal agent that understands you, your goals, and everything you care about to help you improve across those key areas. The tech giant is also focused on building AI that promotes creativity with its tools, inspires business owners to embrace entrepreneurial ideas, improves teaching through personalized tutors & coaches with a PhD in every subject imaginable and pushes for scientific advances.
• Pushing for America to compete with its AI rivals and Meta helping communities: Rather than slowing the development of AI data labs, Zuckerberg is a stern supporter of a world where America must build energy capacity and data centers faster in a bid to compete with China. He also points to Meta launching a “Future Is For Everyone Fund” that’s equipped to financially support the communities that house their data centers. “America's Workforce Academy” is also in place to train those communities’ skilled tradespeople to get access to free training and high-paying jobs in the areas where data centers have been built by Meta.

Final thoughts

Zuckerberg’s extensive essay supports the idea of everyone having easy access to their own personal superintelligent AI, with its presence being used to boost their personal empowerment through several avenues of their lives and a proper balance of power among the biggest entities using AI so that a foundation of safety is instituted.

His concept of AI companies working closely with governments on tighter security measures by sharing their progress before their AI agents are fully deployed sticks out as a movement I can get behind.

More from Tom’s Guide

• 'Raising the standard': Inside Talkspace's bold new AI mental health support tool
• OpenAI’s head of ethics just quit — here’s why ChatGPT users should pay attention
• OpenClaw is available on iPhone and Android — these are the things I would set up first

And while that's not necessarily a bad thing — after all, Meta AI was built specifically for the glasses and can use their camera to answer questions about whatever I'm looking at. But ChatGPT is the AI assistant I use far more often, so I started wondering what would happen if I could give ChatGPT the same view.

And this experiment may offer a glimpse at where ChatGPT is headed anyway. OpenAI is already working with former Apple design chief Jony Ive on a new generation of AI hardware. Its first device is reportedly a portable, screenless assistant equipped with cameras and sensors that can understand what's happening around it, with several more devices said to be in development.

We don't know whether smart glasses will be one of them yet, but the direction highlights that OpenAI is thinking about a future where using ChatGPT doesn't necessarily mean pulling out your phone and opening an app.

I didn't want to wait for that, and with a rumored $300 price tag, I figured I might as well use what I already have: my Ray-Ban Meta glasses.

How I got ChatGPT Vision on my Ray-Bans

To be clear, I didn't jailbreak my Ray-Bans or somehow replace Meta AI. Instead, I used my iPhone as a bridge between the camera on my glasses and ChatGPT.

The result was essentially ChatGPT with a pair of eyes. There are really two parts to making this work: voice and vision. For obvious reasons, voice is the easier half, and my main reason for wanting to do this experiment. I prefer the new ChatGPT Voice.

Ray-Ban Meta glasses connect to an iPhone over Bluetooth and can already handle audio from calls, music and other apps. That means I can run ChatGPT on my iPhone while using the glasses to hear its responses.

Vision takes a little more work. Meta's Wearables Device Access Toolkit gives developers access to features on compatible Meta glasses, including their first-person camera. That creates a way for an iPhone app to receive images from the glasses rather than relying on the iPhone's own camera.

My iPhone effectively became the middleman.

The setup looked something like this:

Ray-Ban camera > iPhone > ChatGPT > Ray-Ban speakers

When I wanted ChatGPT to see something, the app could capture an image from the glasses' point of view and send it along with my question to ChatGPT. ChatGPT could then analyze what I was looking at and respond through the glasses. It's slower, no doubt, and the iPhone is doing a lot of work behind the scenes, but it gets the job done.

Let me be extra clear here: I didn't install ChatGPT directly on the Ray-Bans, and saying "Hey Meta" doesn't suddenly summon ChatGPT. But once the phone was back in my pocket, it started to feel surprisingly close to having ChatGPT glasses.

I asked the same questions to both AIs

Using ChatGPT on a phone requires a deliberate action. I have to take out my phone, open the app and point the camera at something before I can ask about it.

With glasses, the camera is already looking wherever I'm looking. Instead of taking a picture of something I didn't recognize, I could simply look at it and ask about it. I could look at a product and ask ChatGPT what it thought. I could have it read something in front of me and then ask a follow-up question.

1. The grocery store test

My first test was something I've done before with Ray-Ban Meta glasses and ChatGPT seperately, but putting them together was interesting. I started by taking two similar products (Pringles and Cape Cod Potato Chips) and asking both AIs to help me choose between them.

I've used the Meta Ray-Bans with food before, but I wanted to see what ChatGPT could do. They both gave me the same answer but with slightly different presentation. Meta added "if you want..." while ChatGPT was more definitive. If I'm asking for advice, I want a response with clear direction.

2. The 'what am I looking at?' test

This test really showed the difference between Meta AI and ChatGPT in an unexpected way. I showed both AIs my chonky orange tabby cat, but ChatGPT already knew his name, Jolly. Because of ChatGPT Memory already locked in from hundreds of queries, ChatGPT knows more about me than Meta AI.

3. The reading test

ChatGPT's voice and responses are much more friendlier and feel like I'm talking to someone who actually knows me. Meta typically gets straight to the point and offers zero fluff within the conversation.

That's a personal preference, for sure. But for me, I much prefer ChatGPT for even the most boring tasks.

Meta AI has one enormous advantage

There's no getting around this: Meta AI belongs on these glasses. The integration is cleaner because Meta controls everything.

The camera, microphones, speakers, wake word and AI were designed to work together. I can say "Hey Meta," ask something about what I'm seeing and get an answer without thinking about what is happening behind the scenes.

My ChatGPT setup has an iPhone sitting invisibly in the middle of the process.

That introduces more steps and more opportunities for something to go wrong. There can be latency issues and the connection can be finicky. There are things Meta can do with its own hardware that my experiment simply can't reproduce.

If all I cared about was getting the fastest possible answer to "What am I looking at?", Meta AI would be very difficult to beat.

But that's not really why I wanted ChatGPT on my glasses. ChatGPT gave me something different with an overall friendlier vibe. Have ChatGPT as a wearable and not on my phone was a game changer to the point I wish I had tried this much sooner.

The real takeaway

The bottom line for me with this experiment is that I finally got to experience what glasses with a different AI felt like. We're moving toward a world filled with AI hardware: glasses, earbuds, watches, pins and entirely new categories of devices.

OpenAI's work with Jony Ive suggests it is thinking seriously about what ChatGPT looks like when it escapes the screen. Meta is already there with Ray-Ban Meta.

But there's another question I think is going to become important: Why should buying a piece of hardware determine which AI assistant I have to use? I'd love to open the settings for my glasses someday and be able to default to any given AI assistant. I'd even like to see Alexa+ on glasses!

And while that's not how Ray-Ban Meta works today, and my setup isn't nearly as seamless as choosing ChatGPT from a menu, having a choice at all feels like the next step.

More from Tom's Guide

• OpenAI’s head of ethics just quit — here’s why ChatGPT users should pay attention
• I found ChatGPT's secret menu — these 9 features change everything
• Claude has 4 built-in skills I use constantly — here’s how to find them