For someone like me who keeps a close eye on AI every day, it’s easy to see why AI skeptics are sitting back while wearing the “I told you so…” expression.
Far too many instances of hackers using AI to power their cyberattacks, terrifying cases of autonomous AI agents going rogue to bypass internal/external security protocols at prominent companies and a flood of backlash from residents against the AI data centers taking up space in their communities have painted AI in an even worse light.
When you add all those cases to all the primary arguments AI doomers bring up, there’s enough evidence from their end to be convinced of just how harmful AI can be.
I wanted to know what those recurring beliefs from AI doomers are, while also seeing what sort of rebuttal an actual AI tool can offer in response to those statements. I chose ChatGPT for this latest experiment and walked away with a deeper understanding of the arguments brought forth by AI doomers and the opposing responses brought to each one.
1. “A superintelligent AI could make humans powerless.”
I’m used to hearing that an AI model that outperforms humans in the areas of science, strategy and persuasion could easily outwit all the safeguards we place upon it. And at some point, our dependence on something so powerful could lead to a situation where we’ll no longer be able to control it.
ChatGPT’s response pointed out that scenario is not all that plausible since there are a bunch of real-world blockades that could prevent it:
Superior intelligence doesn’t automatically confer unlimited power. A takeover would also require access to computing resources, infrastructure, money, and ways to act in the physical world. Each is a potential barrier. The argument establishes a plausible danger, but it needs more evidence to establish that domination is likely. The 2026 International AI Safety Report says evidence remains insufficient to reliably determine how current capabilities would translate into future loss of control.
2. “AI doesn’t need to hate us—it just needs the wrong goal.”
With the rise of AI agents comes the possibility of them pursuing an objective relentlessly and possibly causing a bit (or even a lot!) of harm to humanity in the process. Rogue AI incidents have sprung up as of late, which lends some credence to the belief that even a harmless command given to an AI system could turn into something dangerous when it's interpreted in a literal sense and pursued at a massive scale.
ChatGPT noted that human intervention in the form of three core safeguards could prevent that scenario from happening:
The most alarming versions assume we give one system a rigid objective, broad authority, and inadequate constraints. That is a deployment choice, rather than an unavoidable property of intelligence. Restricting permissions, limiting the scope of tasks, and requiring approval for consequential actions could interrupt that chain. The unresolved challenge is making those restrictions dependable as capabilities grow.
3. “AI could pretend to be safe until it can escape oversight.”
Another recurring AI doomer argument that has been making the rounds centers around a capable AI model concealing its dangerous intentions before it is deployed by its developers. There’s a bunch of folks out there who feel that the same model could behave well during its testing phase to make its developers trust it, then cause extreme harm as soon as it’s free from its handlers.
ChatGPT argued against that point by bringing up a recent study released by Anthropic that covered that very same topic (which is referred to as “alignment faking”):
Demonstrating deception in a constructed setting doesn’t establish that deployed systems have a persistent plan to overthrow their operators. Anthropic explicitly cautioned that its study did not demonstrate the development of malicious goals. My conclusion is that we shouldn’t trust a model’s assurances about itself: we need independent monitoring and controls. That’s a serious engineering requirement, but it isn’t proof of inevitable betrayal.
4. “AI will give small groups the power to cause mass destruction.”
AI doomers have pointed out that the technology could lower the barrier of expertise needed for hackers to commit cyberattacks or create devastating biological weapons. That worst-case scenario could lead to a world where those types of harmful attacks could be carried out by smaller groups over time as they get advice on how to carry out mass destruction from AI models.
ChatGPT brought up the factors that could impede that scenario from coming true, plus it brought up a report that mentions the preventive efforts that could do the same:
Access to advice is only one part of carrying out an attack. Equipment, materials, execution, and avoiding detection can remain substantial obstacles. AI can also strengthen defenses, although we shouldn’t assume defenders will win. The safety report identifies uncertainty about real-world biological misuse and whether AI benefits cyberattackers or defenders more. This supports targeted prevention, rather than certainty that catastrophe follows from access.
5. “Competition will force companies and governments to sacrifice safety.”
It has been disappointing to see how companies such as OpenAI, Google, Meta and Microsoft have cut corners when it comes to data collection, safety testing and product readiness to gain the number one spot in the AI race. If this trend continues, we could reach a point where all of their risky efforts could lead to an AI catastrophe that we all know none of them collectively want to see come to fruition.
ChatGPT’s counterargument points out how efforts put in place to hold those companies accountable for cutting corners could prevent a future disaster from occurring. While I come across plenty of lawsuits and proposed government regulation brought against AI companies, it’d be nice to see stronger laws put in place to punish them for sacrificing safety for progress:
Competitive incentives are powerful, but they aren’t fixed laws. Enforceable standards, independent scrutiny, and consequences for preventable harm could change the calculation. The weakness in the fatalistic version is treating every attempt at coordination as doomed. Still, this rebuttal depends on building effective institutions; voluntary promises alone give little reassurance.
6. “Mass automation could destroy livelihoods and concentrate power permanently.”
It’s been pretty discouraging to see job losses become even more prevalent, with one of the determining factors being companies placing all their bets on AI being able to automate some of their workers’ tasks. AI doomers argue that as AI evolves and becomes more capable of performing economically viable work for major corporations, workers could lose their bargaining power as those companies become more reliant on AI to fill their roles.
ChatGPT opposed that argument by stating that AI won’t get rid of several occupations just cause it can automate a few tasks:
Automating tasks doesn’t translate directly into eliminating entire occupations, and higher productivity could create new demand. But history cannot guarantee a painless transition. My strongest objection is to treating unequal outcomes as technologically predetermined: ownership, competition, taxation, and public services would help shape who benefits. Economists disagree substantially about the eventual employment effects.
7. “AI-generated persuasion and surveillance could make authoritarianism unbeatable.”
We’re already seeing how AI is being used to politically persuade the masses via AI-generated photos and videos. The presence of fabricated evidence of supposed politically fueled acts that are deemed negative and AI-powered surveillance could get worse over time. Governments and corporations could take advantage of AI-made propaganda to manipulate their populations even more and suppress their opposition as the tech continues to evolve.
ChatGPT countered that argument by stating humans aren’t so easily susceptible to AI propaganda and that AI could even be used to fight back against government/corporations:
Producing persuasive content cheaply doesn’t guarantee control over what people believe or do. Audiences, institutions, and competing sources of information can resist manipulation. AI could also help expose corruption and analyze evidence. The danger is credible, but the claim that authoritarian control becomes permanent assumes those counterforces all fail.
The takeaway
I totally get both sides of the aisle who argue for and against AI—the AI doomers have seen some of their scenarios start to come to life, while the people who argue against those same points present credible rebuttals. ChatGPT highlighted highly researched reports and even its own counterarguments to present a more optimistic viewpoint of what’s currently happening with AI.
The chatbot’s final statement did a good job of pointing out a viewpoint that meets both sides in the middle of the aisle:
These arguments are strongest as reasons to reduce specific risks. They become much weaker when they turn uncertain chains of events into confident predictions that humanity is doomed. Equally, uncertainty gives us no basis for confidently dismissing them.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom’s Guide
- AI is changing cancer treatment — and a new breakthrough shows just how far we've come
- I compared OpenAI Codex and Claude Code's usage limits — and there's a catch with both $20 plans
- These are the 7 best ways to use ChatGPT before you buy anything on Black Friday, from choosing what’s actually worth buying to auditing your final purchases
