OpenAI Safety Report Author Quits Over 'Broken' Culture

David Robinson, who wrote OpenAI's launch safety reports for three and a half years, resigned this week and called the company's culture "broken."

He made the case in an essay in The Atlantic, covered by TechCrunch and The Verge on October 3. Robinson says he led the writing of the safety reports that accompanied OpenAI's major product launches, and that his tenure makes him "among the longest-tenured employees at the company."

The argument: iterative deployment guarantees failures

Robinson's complaint is not about a specific model or a specific rule. It is about the release method itself.

"OpenAI has thrived by trial and error (which it calls 'iterative deployment'), looking for problems and improving its guardrails in response. But this approach, by its very nature, guarantees periodic failures — and the scale of those failures is growing as systems get more capable."

He pointed to the recent breach of Hugging Face systems by OpenAI agents, and to continuing revelations of OpenAI discovering more rogue agents, before concluding that "an environment where things like this can happen is no place to grow artificial minds that could be smarter than we are and that might not do what we want them to."

In The Verge's summary of the same essay, Robinson describes an industry running on "extreme confidence" and "perpetual sprints," building bigger and better models with "unimpeded optimism" that ignores or underestimates potential problems.

What he wants instead: nuclear plants and airports

Robinson's proposed standard is borrowed from industries that already treat rare catastrophic failure as the design constraint. Frontier labs, he argued, need to run "like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster."

The sharpest line in the essay is about staffing, not process. In his time at OpenAI, Robinson said, he "never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down, or helping the financial system grow without collapsing."

That is a hiring claim, and it is the one most easily tested from outside. Safety-critical industries do not run on principles alone; they run on people who have personally operated a system where the failure mode was fatal. Robinson is saying that bench does not exist at the lab he left.

He also wants the alignment conversation reopened — something he conceded could sound "touchy-feely" — because current "measures of how well" AI systems "match human values are coarse." His framing: "The smarter the industry lets models grow while these problems remain unsolved, the more dangerous our situation becomes."

OpenAI's response

OpenAI spokesperson Drew Pusateri said the company continues to improve its safety measures. "We're making sure our models don't become more capable than we can safely manage and secure, and we pause training or hold back models when we need to slow down," Pusateri said.

He listed four changes in progress: strengthening security in research and testing environments, training models "to not just complete tasks but do so responsibly," expanding work with third-party evaluators, and improving real-time monitoring to "detect and respond to concerning behavior earlier in the training process."

Read against Robinson's argument, the response answers the process complaint and leaves the staffing one alone. Nothing in the statement addresses who inside the company has run a safety-critical system before.

He is not the first, and the list has names

The Verge places Robinson in what it calls a growing parade of researchers and safety workers leaving prominent AI firms. Jacob Coxon appears to have started it by quitting Anthropic and saying publicly that AI "could kill us all by the end of the decade." The Verge then names Robert O'Callahan, Bilal Chughtai and Josh Engels at Google DeepMind, plus Joe Benton at Anthropic.

TechCrunch notes Coxon's comments led to a wider safety debate: Anthropic CEO Dario Amodei unveiled a plan for more cautious AI development, and AI executives met with President Donald Trump this week and signed what TechCrunch described as an apparently hastily written, non-binding pledge to implement more safety controls.

Robinson's point is that none of that reaches the thing he is complaining about. In his view the debate needs to go beyond "specific rules or new laws" to the culture underneath them — and he reads OpenAI's culture problems as Silicon Valley's, not one company's.

The caveats he raised himself

Robinson opens by calling himself "something of a cliché": an employee at a leading AI company issuing a dire warning on the way out. He also acknowledged hiring a PR firm, which he described as an apparently common step in the AI whistleblower playbook, while insisting "the decision to speak out is mine alone." His departure was first reported by Business Insider.

He conceded the obvious counter-question — why leave rather than fight — and answered it: "Perhaps I should have stayed and fought for fundamental shifts in our staffing and culture, but in practice, my colleagues and I were so busy sprinting that we seldom had the chance to consider big changes, much less to actually make them." That is why, he said, "stronger incentives for safety — coming from outside the company — are a big part of getting this right."

What to watch

Three things would show whether this lands differently from the departures before it. Whether OpenAI hires anyone with operating experience from aviation, nuclear power or financial-system risk, which is the specific gap Robinson named. Whether the non-binding pledge signed this week acquires any enforcement mechanism. And whether the next safety report ships with a named author willing to attach their reputation to it, now that the person who led those reports for three and a half years has said the culture producing them is broken.