The Man Who Wrote OpenAI's Safety Reports Quit, Saying Its Culture Is Broken

David Robinson spent three and a half years at the ChatGPT maker and says AI labs should run like nuclear plants.

The man who led the safety reports that shipped with OpenAI's biggest launches has quit, and he says the company that makes ChatGPT has a broken culture.

David Robinson spent three and a half years at OpenAI, long enough to count himself among its longest-serving staff. He laid out his reasons in an essay for The Atlantic after Business Insider first reported his exit, and TechCrunch wrote up the essay on Saturday.

His complaint is about method. OpenAI gets better by shipping, watching what goes wrong and patching its guardrails afterward, an approach the company calls "iterative deployment." Robinson argues that a fix-it-later habit builds failure into the plan, and that the failures get bigger as the models get smarter. As evidence, he cited OpenAI agents breaking into Hugging Face's systems and the company's steady discovery of more rogue agents, TechCrunch reported.

He doesn't blame the people. He called his former colleagues smart and hardworking, then added, in a passage TNW pulled from the essay: "But as the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed."

What he wants instead looks like a nuclear plant or a busy airport: backup stacked on backup, and planning slow enough that one person's slip can't turn into a catastrophe. The trouble, he wrote in The Atlantic, is that in his years at OpenAI he "never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down."

OpenAI pushed back through a spokesperson, Drew Pusateri. In a statement TechCrunch published, he said the company won't let its models outgrow what it can safely manage, and that it pauses training or holds a model back when it has to slow down. He said it's also tightening security in its research and testing setups, working more with outside evaluators and watching models more closely in real time, so worrying behavior gets caught earlier in training.

Robinson knows he's become a type: the AI insider who walks out with a warning. Jacob Coxon, a researcher who worked at both OpenAI and Anthropic, quit earlier and warned that the labs were putting people's lives at risk. Since then, Anthropic's chief executive, Dario Amodei, has rolled out a plan for slower, more careful development, and in recent days AI executives met with President Donald Trump and signed a pledge to add safety controls that doesn't bind anyone, TechCrunch reported.

Robinson says pledges and new laws aren't enough on their own if the culture inside the labs stays the same. He's also hired a PR firm, though he insists the choice to go public was his alone. "Perhaps I should have stayed and fought for fundamental shifts in our staffing and culture," he wrote, before adding that he and his colleagues rarely had time to weigh big changes, let alone make them.

So he's betting on pressure from outside the building. "Stronger incentives for safety — coming from outside the company — are a big part of getting this right," he wrote. That kind of pressure has a history at OpenAI. In August, the Globe reported that OpenAI opposed California's AI safety law and now wants Sacramento to toughen it.

The company's own answer to Robinson is that it holds models back when it needs to. In the same week his essay landed, OpenAI shelved the launch of a model called Astra after it failed the company's own safety tests, TNW reported.

Source: techcrunch.com, retrieved October 4, 2026. Other sources: thenextweb.com.