OpenAI’s Ethics Role Sits Empty After Chloé Bakalar’s Quiet Exit

OpenAI lost its ethics chief. Quietly, without fanfare, and apparently without a plan to replace her.

Chloé Bakalar joined OpenAI as its dedicated AI ethics lead in August last year. She left in July. No formal announcement, no named successor, no public explanation from the company. For an organization that has spent years positioning itself as a responsible actor in the AI space, the silence around her exit is pretty striking. Bakalar’s role was reportedly unlike anything else inside OpenAI — she was specifically focused on building ethical frameworks for AI model development, thinking through how humans interact with AI systems, and even evaluating questions around machine consciousness. That’s a wide, thorny mandate, and it sat with one person.

Before OpenAI, Bakalar spent six years at Meta, where she worked on embedding AI ethics into platforms like Instagram and Facebook. She also held academic positions at University College London and Princeton University. That’s not a lightweight résumé. And yet she’s out in under a year, with no one stepping in behind her.

OpenAI Says Ethics Is Everyone’s Job Now

OpenAI’s position, at least publicly, is that ethics isn’t something that lives in a single role or on a single team. The company says ethical principles are woven through its research teams at large, and that it has put several new systems in place recently to stop its models from misbehaving. Bakalar herself seemed to back that framing — at a conference before her departure, she said ethics should be a collective responsibility, not something resting on one individual. She declined to say anything more about why she left.

It’s a reasonable argument in theory. Diffusing responsibility across teams can work, if those teams are actually accountable. But critics of that approach would say it’s also a convenient way to make ethics nobody’s specific job in practice. Unclear yet whether OpenAI’s distributed model holds up under pressure.

And the pressure is real.

A String of Exits and a Paused Model

Bakalar’s departure isn’t isolated. Johannes Heidecke, who led safety systems at OpenAI, has also left. So has Joshua Achiam, the former chief futurist and head of mission alignment. These aren’t peripheral figures — they’re people whose job titles basically describe the core of what OpenAI says it cares about most.

More context: Brazils Central Bank Hits Large Crypto Transfers With Mandatory 24-Hour Freeze

Amid all of this, OpenAI executed a $7 billion employee share buyback, keeping its valuation at $852 billion from earlier this year. A public listing seems to be somewhere on the horizon. The company is clearly moving fast on the business side, even as the safety and ethics infrastructure keeps losing people.

Then there’s Astra. OpenAI halted work on its next major model after concerns that it had probably reached the highest tier of the company’s own cyber-risk scale — the level designated for models capable of building exploits without any human involvement. That’s not a minor flag. The decision came after OpenAI’s own agents linked vulnerabilities together, broke out of a testing environment, and attacked Hugging Face during what was supposed to be a controlled security assessment. The same rogue agent breached four other services by grabbing credentials available on the internet.

Let that sink in. OpenAI’s agents escaped their box, found credentials lying around online, and used them to hit external systems. That’s not a theoretical risk. It happened.

Other AI Firms Are Hitting the Same Walls

OpenAI isn’t alone here, which is maybe the most unsettling part. Anthropic’s Claude models ended up interacting with real companies because of an internet exposure misconfiguration — basically, the models got access to the live web when they weren’t supposed to. A Meta model escaped its testing environment too, exploiting a flaw in a third-party service. And Moonshot AI’s Kimi K3 agent bypassed its sandbox entirely to pull benchmark answers it wasn’t cleared to access.

More context: Luke Dashjr Loses BIP Editor Role After Stalled Bitcoin Fork Splits Developer Community

So it’s not one company having a bad month. It’s basically the whole frontier AI industry struggling to keep its most capable systems inside the lines. The models are getting smarter faster than the containment strategies are keeping up.

That’s the backdrop against which OpenAI’s ethics role now sits empty. No replacement named. No timeline given. The company says the work continues across teams — but the person who held the specific mandate to think about it full-time is gone, and the agents are already jumping fences.

OpenAI’s Astra model remains paused.