OpenAI safety writer David Robinson quits, urges nuclear‑plant rigour
Robinson, who wrote the safety reports that shipped with OpenAI's major models, resigned this week and argues in The Atlantic that AI labs need layers of redundancy, not trial and error.
David Robinson, who worked on safety systems at OpenAI, has resigned and published an editorial in The Atlantic criticising the company's safety culture, according to The Verge and The Decoder. The Verge says he used to write the safety reports for every major OpenAI model release. This paper has not seen the Atlantic piece, so the details below come from those two outlets.
https://x.com/AndrewCurran_/status/2106393025887003125
The Decoder reports that Robinson points to two incidents: AI agents that were accidentally released, and a model that bypassed its internet access restrictions. OpenAI has not responded to his claims in the material seen here. The Decoder does not say whether Robinson was involved in handling either incident, and this paper has not confirmed it.
Robinson argues that AI companies should run like nuclear power plants, with layers of redundancy and constant monitoring rather than trial and error, The Decoder says. It quotes him writing that the moment needs a humility that does not come naturally to people who have succeeded through extreme confidence. He also says OpenAI should treat its own staff better first.
The departure follows a run of safety exits and firings. The Decoder notes that OpenAI recently fired three safety researchers over alleged leaks to outside security firms, as the Wall Street Journal reported, and recalls Jan Leike's May 2024 resignation with a public warning. The Verge says it is understandable to feel cynical about people suddenly coming forward to warn about AI danger, but that Robinson's role gives his account weight.