OpenAI safety lead David Robinson resigns over a ‘broken’ culture
David Robinson, who led the writing of safety reports for OpenAI's major launches, has resigned with an essay arguing that trial-and-error safety cannot keep pace with the models it ships.
An OpenAI employee who led the writing of the safety reports that accompanied the company's major product launches has resigned, arguing that its approach to safety is structurally unable to keep pace with the systems it ships.
‘Trial and error’
David Robinson published his argument as an essay in The Atlantic on 3 October 2026. He said he had spent three and a half years at OpenAI, making him “among the longest-tenured employees at the company”, and concluded that its “culture is broken”.
The case he makes is about process rather than any single rule. “OpenAI has thrived by trial and error (which it calls ‘iterative deployment’), looking for problems and improving its guardrails in response,” he wrote. “But this approach, by its very nature, guarantees periodic failures — and the scale of those failures is growing as systems get more capable.”
Not the first exit, and not only OpenAI
Robinson's departure follows that of Jacob Coxon, a researcher who worked at both OpenAI and Anthropic before leaving and describing the labs as “gambling with our lives”. Coxon's intervention pushed AI safety up the agenda: Anthropic chief executive Dario Amodei set out a plan for pacing frontier development, and days earlier AI executives met President Donald Trump and signed what appeared to be a hastily written, non-binding pledge to implement more safety controls.
Where Robinson departs from the others is in scope. His essay locates the problem in the wider culture of the industry rather than in one company's decisions, arguing that the debate needs to move past specific rules and laws towards how these firms behave day to day. Coverage of the resignation ran at Reuters, Bloomberg, The Guardian and elsewhere the same day, with several outlets leading on his call for safety oversight modelled on industries such as aviation, where incidents are investigated and the findings fed back into design.
Our opinion
The interesting thing about this resignation is how unremarkable its shape has become. A senior safety figure leaves, writes that the culture is broken, and the argument lands as one more entry in a genre that now has its own rhythm — which is precisely the problem Robinson identifies. His complaint is not that OpenAI ignores safety but that it treats safety as a beta feature, patched after the fact and measured by the absence of incidents until an incident arrives. The aviation comparison is the sharpest part, because it points at the missing ingredient: an industry that learns from failure in public, with findings that bind the next design. AI labs do the opposite, letting each near-miss stay a private memo. A resignation essay cannot fix that, but it can at least make the pattern hard to ignore.