David Robinson quits OpenAI, calling its culture 'broken'

Robinson is not an outside critic. He spent his time at OpenAI writing the safety reports that accompanied every major model release, which makes his resignation harder to dismiss as the complaint of someone unfamiliar with how the lab actually works. His editorial in The Atlantic argues that the problem is not a missing rule but a culture: Silicon Valley, he says, operates with 'extreme confidence' and 'perpetual sprints,' building bigger and better models under 'unimpeded optimism' that ignores or underestimates potential problems. That framing is why his proposed fix is structural rather than procedural — he says frontier labs should run like nuclear power plants or busy airports, with layers of redundancy and careful, time-consuming planning so the occasional, inevitable human error does not open a door to disaster.

He is also not alone. Robinson joins a growing group of researchers and safety workers who have left prominent AI firms, a wave that appears to have started with Jacob Coxon, who quit Anthropic and publicly said AI 'could kill us all by the end of the decade,' followed by Robert O'Callahan, Bilal Chughtai, and Josh Engels at Google DeepMind, and Joe Benton at Anthropic. Together they turn what could read as one person's grievance into a pattern that is harder for the labs to wave away.

The underlying tension the piece raises is between speed and prudence. The question is whether labs treat these departures as a reputational problem to manage or as evidence that the release process itself needs to be slower and more layered, as Robinson suggests. That is likely to hinge on whether more staff follow the same route — and whether the companies adjust anything before they do.