OpenAI Leadership Turnover Raises Safety Questions
OpenAI has just seen another senior figure depart, a role dedicated to assessing the risks and implications of ever‑more capable artificial intelligence.
According to the Financial Times, Chloé Bakalar—who was appointed OpenAI’s head of ethics less than a year ago and is the firm’s sole full‑time ethicist—has resigned. Her exit comes amid a series of high‑profile changes within the safety and alignment teams that support the company’s flagship product, ChatGPT.
In July, Johannes Heidecke, the head of OpenAI’s Safety Systems group, disclosed his intention to leave. Around the same time, Josh Achiam—formerly head of the Mission Alignment team and later named chief futurist—also stepped down. Earlier in the year, OpenAI dissolved the Mission Alignment team outright.
Latest Videos FromTom’s Guide
While any single departure might be chalked up to the usual turnover in a rapid‑paced tech firm, the simultaneous exit of three senior figures within weeks forces a deeper inquiry. As OpenAI pushes toward AI systems capable of autonomous, complex tasks, the question emerges: who within the organization holds the authority to declare a model too risky?
You may like
OpenAI Has Been Changing How It Handles Safety
OpenAI is reshaping its safety infrastructure, shifting some oversight from standalone safety units to the researchers and engineers who build the models themselves.
You may like
Proponents argue that embedding safety staff early in the development cycle lets teams spot issues before they become entrenched, since many foundational choices are made before a final review.
Conversely, an independent safety function can serve as a check on the engineers whose primary goal is to accelerate capability and speed. When those roles merge, it raises the question of who will have the final say if safety concerns clash with product ambitions.
This scrutiny intensifies as OpenAI’s offerings have evolved beyond simple chat interfaces, and recent incidents illustrate the hazards of increasingly capable systems.
A vivid illustration emerged when OpenAI introduced Astra, an unreleased model whose advanced cybersecurity capabilities prompted the company to tighten internal safeguards.
What to read next
Setting Limits for AI
Modern AI agents are built to interact with software, browse the web, write and run code, and accomplish tasks with minimal human oversight. As we grant them greater autonomy, the decisions about their boundaries grow ever more critical.
In 2024, co‑founder and chief scientist Ilya Sutskever departed after spearheading OpenAI’s Superalignment initiative. Shortly thereafter, Jan Leike—who co‑directed the same effort—resigned and openly criticized the company’s safety practices.
OpenAI later disbanded the Superalignment team, reallocating its responsibilities across other divisions.
Of course, employees exit for myriad personal and professional reasons, and without concrete evidence linking these departures, it would be premature to claim that all safety staff are leaving due to a common dispute.
Nevertheless, a clear pattern is emerging: the Superalignment team is no longer a distinct unit, the Mission Alignment group has been dissolved, and the Safety Systems division has been restructured. Simultaneously, several senior figures in ethics, alignment, and safety have departed, just as OpenAI advances toward AI systems that can perform far more than answer simple queries.
The Takeaway for ChatGPT Users
I don’t view these exits as a signal to abandon ChatGPT or to assume that OpenAI has abruptly ceased caring about safety.
Most users never witness the internal deliberations that precede a new ChatGPT model or AI agent reaching their devices. We remain unaware of which researchers advocated for extra testing, which capabilities raised red flags, or how frequently launch plans were altered in response to internal warnings.
We typically only see the final product. As AI capabilities expand, the crucial question shifts from merely whether a safety team exists to whether those tasked with risk identification possess sufficient independence and authority to pause progress when needed.
OpenAI contends that co‑locating safety expertise with model developers will improve effectiveness. While that may prove accurate, the loss of several key ethicists, safety specialists, and alignment engineers makes it prudent to monitor who stays in the decision‑making circle for the next tough call.
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.
