OpenAI's Ethics Team Just Went to Zero

One person, one job, then nobody

Chloé Bakalar joined OpenAI in August 2025 as its dedicated ethics lead, arriving from nearly six years as chief ethicist at Meta, where she worked on ethics review for AI features across Facebook and Instagram and held academic appointments at Princeton, UCL and Temple. At OpenAI her mandate covered ethical approaches to model development, how people interact with AI systems and the harder edge-case questions around machine consciousness. In July 2026, according to press reporting that surfaced weeks after the fact, she left. OpenAI did not announce it.

The company has not filled the role. Reporting across multiple outlets describes Bakalar as OpenAI's sole dedicated ethicist, meaning her departure removed the only person doing the job, full stop. She joins a list that already includes Johannes Heidecke, head of safety systems, and Joshua Achiam, who led mission alignment work for nearly nine years before leaving this year. Three safety-adjacent exits in one year, at a company running an unusually fast release cadence, is a pattern worth reading carefully rather than a single resignation to shrug off.

What OpenAI said instead of an explanation

Asked about the departure, OpenAI did not dispute the facts and did not name a successor. It offered a different frame: ethics, the company said, does not live with one owner or team, and ethical considerations are embedded into model building by a number of teams across research. That statement may be true as an operating description. It is also the sentence every organization reaches for the moment a dedicated function disappears, because it converts an absence into a design choice.

The useful test checks whether the work described is still visible anywhere on an org chart, in a budget line, or in a single person's job title. Right now none of those exist at OpenAI for ethics specifically. A responsibility distributed across every team with none of them accountable for it becomes, in practice, a responsibility nobody owns - which is the arrangement OpenAI is now asking the public to trust runs itself.

Read the resourcing, not the mission statement

Whether OpenAI's models are safe, or whether the company acted in bad faith, are claims this single departure cannot settle either way. The more useful signal is one every owner-operator can read in their own organization: what a company funds during a growth sprint is a more honest statement of its priorities than what it says in a press line. OpenAI is shipping models faster than at any point in its history and raising capital at a pace that dwarfs its research budget of three years ago. Against that backdrop, a single-person ethics function was always going to be the cheapest line to cut, and it was cut.

Owners run smaller versions of this test constantly without noticing. A compliance hire who leaves and is never backfilled. A security lead whose duties get split across three engineers already at capacity. A customer-trust role folded into marketing the week after a funding round closes. In each case the official language stays warm - embedded across the team, everyone owns quality now - while the actual answer sits in a budget line that quietly went to zero. Watching that line beats reading the memo.

The audit you can run this week

Take this beyond OpenAI's specific case and turn it into a check on your own operation. List every function in your business that currently depends on exactly one person - ethics, compliance, security, quality, whatever the equivalent is for your industry. For each one, write down what happens on the day that person resigns: does the role get a backfill commitment within a fixed window, or does the work get silently redistributed to people who already have full-time jobs? Decide the answer now, while it is still a calm policy choice rather than a live scramble.

The useful discipline is treating headcount-of-one functions as a single point of failure the same way you would treat a single server with no redundancy, and pricing the replacement cost into your plan before the vacancy happens rather than after. OpenAI's ethics function went from one person to zero people in a single departure, with no public plan for what replaces it. Whatever you conclude about OpenAI, the diagnostic transfers cleanly to any organization growing faster than it is staffing the functions that do not generate revenue directly.

Read next: Hugging Face Had To Ask For Its Own Breach Logs  |  Microsoft Now Caps What Its Own Engineers Spend on AI