OpenAI has lost another senior leader whose job involved thinking about the risks and consequences of increasingly powerful AI.
As reported by Financial Times, Chloé Bakalar, OpenAI’s head of ethics and reportedly its only dedicated ethicist, has left the company less than a year after joining. Her departure follows other significant changes among the people responsible for safety and alignment at the company behind ChatGPT.
Johannes Heidecke, who led OpenAI’s Safety Systems team, announced his departure in July. Josh Achiam, who previously led OpenAI’s Mission Alignment team before becoming chief futurist, has also reportedly left the company. OpenAI had already disbanded the Mission Alignment team earlier this year.
Any one of those changes could easily be written off as the normal churn of a fast-moving technology company. But when three leaders leave all within the matter of weeks, it raises a bigger question at a moment when OpenAI is building AI that can do complex work on its own: Who inside the company has the power to say a model has gone too far?
OpenAI has been changing how it handles safety
OpenAI has been reorganizing how safety works inside the company, moving some of that responsibility closer to the researchers and engineers actually developing its models rather than keeping safety work confined to separate teams.
There is an argument for doing it this way. A safety team brought in at the end of development can identify problems, but by then many of the decisions that shaped the model have already been made. Putting safety researchers alongside the people building the technology could allow potential problems to be identified much earlier.
But there is another side to that argument. An independent safety team can act as a counterweight to the people whose job is to make a model smarter, faster and more capable. As those responsibilities become intertwined, it becomes reasonable to ask who ultimately gets to push back when the safest decision conflicts with the desire to release a new product.
That question is becoming harder to dismiss because OpenAI’s models aren't just chatbots anymore, and we're already seeing what can happen when AI becomes too capable.
OpenAI recently provided an unusually concrete example of this problem with Astra, an unreleased model whose cybersecurity abilities became powerful enough to trigger additional precautions inside the company.
OpenAI said it could not rule out the possibility that Astra could reach its highest level of concern for cybersecurity capabilities. Rather than immediately releasing the model broadly, the company moved parts of its testing into more isolated environments and began considering stricter controls around who could access its most powerful capabilities.
Setting limits for AI
AI agents are continually designed to use software, browse websites, write and execute code and complete tasks with less human supervision. The more freedom we give these systems, the more important the decisions around their limits become.
That makes the people setting those limits even more important than ever. And, unfortunately, this isn't the first time OpenAI has lost prominent safety researchers.
Now several senior people associated with ethics, alignment and safety have left. This is all happening at the same time OpenAI is building systems capable of doing significantly more than answering questions in a ChatGPT window.
In 2024, co-founder and chief scientist Ilya Sutskever left the company after playing a central role in its Superalignment effort. Jan Leike, who co-led that team, resigned shortly afterward and publicly criticized OpenAI's approach to safety.
OpenAI subsequently dissolved the Superalignment team and distributed its work elsewhere in the company.
Of course, people leave companies for all kinds of reasons, and without evidence connecting their decisions, it would be unfair to suggest that everyone working on safety at OpenAI is heading for the exits because of some shared disagreement.
But the organizational trend is certainly notable. Superalignment disappeared as a standalone team, Mission Alignment was later disbanded and Safety Systems has been reorganized. Now several senior people associated with ethics, alignment and safety have left. This is all happening at the same time OpenAI is building systems capable of doing significantly more than answering questions in a ChatGPT window.
The takeaway for ChatGPT users
I don't think these departures are a reason to stop using ChatGPT or assume that OpenAI has suddenly stopped caring about safety.
Most of us will never see the internal debate that happens before a new ChatGPT model or AI agent reaches our phones and computers. We don't know which researchers argued for additional testing, which capabilities caused concern or how often a launch was changed because somebody inside the company raised a red flag.
We mostly see the result. So as AI becomes more capable, the question isn't simply whether companies such as OpenAI have a "safety team." It's whether the people responsible for identifying risks have enough independence and authority to slow things down when necessary.
OpenAI says bringing safety closer to model development can make that work more effective. That may prove true, but after losing several of the people who were specifically tasked with thinking about ethics, safety and alignment, it's reasonable to pay closer attention to who remains in the room when the next difficult decision has to be made.
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. Finally, you can visit our dedicated Tom's Guide Savings Squad hub for expert help on getting the best products for less.
More from Tom's Guide
- I put ChatGPT on my Ray-Ban Meta glasses — and it completely changed how I used them
- ChatGPT can do more for you than ever — and I use these 5 prompts to take everyday tasks off my plate
- found ChatGPT's secret menu — these 9 features change eveClaude has 4 built-in skills I use constantly — here’s how to find them
