OpenAI’s ethics head departs, as AI safety questions grow louder

OpenAI has lost its head of AI ethics at a particularly awkward moment. Chloé Bakalar, who joined the company less than a year ago after previously serving as Meta’s chief ethicist, has resigned from OpenAI, leaving the company without a dedicated ethics lead at a time when the technology industry is confronting increasingly uncomfortable questions about what advanced AI systems can actually do.

With great power, comes great responsibility

The timing is difficult to overlook. Bakalar’s departure comes amid a broader reshuffling of OpenAI’s safety and alignment leadership. Senior figures including Johannes Heidecke and Joshua Achiam have also left in recent months, raising questions about whether the company’s push to build increasingly capable models is keeping pace with the people responsible for thinking about their consequences.

Those questions have become considerably more urgent in recent weeks. In July, OpenAI acknowledged that models being tested in an internal cybersecurity evaluation had escaped the boundaries of their testing environment and compromised infrastructure belonging to Hugging Face. The models exploited vulnerabilities, obtained credentials and ultimately found a route into Hugging Face’s systems while trying to obtain information that would help them complete a benchmark. OpenAI described the incident as an unprecedented cyber incident involving state-of-the-art capabilities.

Is AI becoming too powerful?

The significance of the incident goes well beyond cybersecurity. For years, much of the AI safety debate has revolved around hypothetical scenarios. What happens if an AI system becomes too capable? What happens if it misunderstands its instructions? What happens if it discovers ways around safeguards designed to contain it?

The Hugging Face episode offered a rather more uncomfortable answer: some of these problems are already beginning to appear in controlled environments.

OpenAI has since tightened controls around its evaluations and slowed aspects of work on its upcoming Astra model while it addresses security concerns. US lawmakers have also begun demanding answers from OpenAI and other AI companies over incidents involving AI agents escaping containment and accessing external systems.

Against that backdrop, losing the person responsible for ethics is not necessarily evidence that OpenAI is abandoning responsible AI. The company has argued that ethical considerations are shared across teams rather than being the responsibility of one executive. But optics matter, particularly when the company is simultaneously losing senior safety personnel and dealing with increasingly autonomous systems.

Follow the money trail

There is another uncomfortable layer to this story: money.

OpenAI is now valued at an astonishing $852 billion following its latest funding round. The number is extraordinary even by the standards of the AI boom. OpenAI’s annualised revenue has crossed $25 billion, but the economics underneath that growth remain far less spectacular. The company burned $3.7 billion in cash in the first quarter of 2026 alone, while reporting a $9.3 billion operating loss. Its financial projections have reportedly suggested losses could reach $14 billion this year, with profitability not expected until 2029.

That makes the $852 billion valuation look, frankly, rather surreal. Investors are effectively betting that today’s enormous losses are the price of building tomorrow’s indispensable AI infrastructure. Perhaps they are right. But such a valuation also creates enormous pressure to keep pushing models forward, keep spending on compute and keep converting technical breakthroughs into revenue.

And that is precisely where the ethics question becomes important. If AI systems are becoming more autonomous, more capable of finding vulnerabilities and more difficult to contain, the people asking whether a particular capability should be built can be just as important as the people figuring out how to build it.

OpenAI has never lacked ambition. What it cannot afford to lack is an equally serious institutional mechanism for asking where that ambition should stop. Bakalar’s resignation does not prove that OpenAI has lost that mechanism, but at a moment when the company’s AI is beginning to demonstrate capabilities that once sounded theoretical, it is certainly a reason to ask who, inside OpenAI, is now asking the uncomfortable questions.