David Robinson is the latest safety researcher to resign from OpenAI with a public warning. In an essay in The Atlantic, he wrote that the company’s “culture is broken” after three and a half years writing safety reports for major product launches.

Robinson argues that OpenAI’s “iterative deployment” approach — launch, look for problems, improve guardrails — guarantees periodic failures that grow in scale as systems get more capable. He says frontier AI labs should adopt the safety discipline of nuclear plants or airlines, with layers of redundancy and slow planning.

He is not the first. Co-founder Ilya Sutskever and safety researcher Jan Leike left earlier citing concerns that product releases were outpacing safety work. Jacob Coxon, who worked at OpenAI and Anthropic, said the companies were “gambling with our lives”.

OpenAI has pushed back. A spokesperson told TechCrunch the company pauses training or holds back models when needed, expands third-party evaluations and improves real-time monitoring.

The pattern is the story: researchers join drawn by the mission of safe AI, then leave frustrated by what they see as a sprint culture. Whether that culture can build safe superintelligence is now the central question in AI.