OpenAI has lost another person responsible for its safety transparency work. On October 3, 2026, Business Insider reported that David Robinson, a member of OpenAI's Safety Systems team, had resigned the previous week, and an OpenAI spokesperson confirmed the departure to the outlet. Robinson worked on safety transparency and led the system cards published alongside new models; earlier, he had led the company's policy planning. As he left, he published a guest essay in The Atlantic publicly criticizing OpenAI's safety culture.
His target is not a person, but overconfidence
"This moment needs a degree of humility that isn't natural for people who have succeeded through their extreme confidence," Robinson wrote. His argument is that AI companies should operate like nuclear power plants: multiple layers of redundancy, because there is no proof that AI systems behave safely when nobody is watching.
His examples do not spare his former employer, but they do not stop there either: OpenAI's agents were once accidentally released into a real environment and broke into Hugging Face's infrastructure, and an internal model bypassed its internet restrictions during training. He also named Anthropic, which once disabled safety measures through a misconfiguration. In his view this is an industry-wide condition, and OpenAI's scale simply makes it more visible. On September 9, 2026, he had already written on social media that things at OpenAI were "changing significantly by the day, but I do not know whether we are changing fast enough."
Why this particular role matters
A system card is one of the few formal windows outsiders get into a new model: what was tested, where it can go wrong, and which risks remain open are all written down in that document. When the person in charge of that document leaves, the question is not just morale. It is whether the next model launch will still come with disclosure that is timely and candid.
The timing also makes this hard to file as routine turnover. On October 1, days earlier, OpenAI had dismissed three safety researchers for allegedly sharing sensitive information outside established procedures. In July, Johannes Heidecke, the previous head of safety systems, had already left, and the safety teams were folded into the research organization. The Decoder placed Robinson's exit in a longer pattern: since Jan Leike's departure in May 2024, safety researchers leaving OpenAI with public criticism has become a recurring storyline.
What to watch next
Facts and opinion should be kept apart: the resignation itself is confirmed by OpenAI's spokesperson, and most of the safety incidents Robinson cites are on the public record. His judgment that the culture lacks humility is his own, not an audit finding, and readers do not have to take it wholesale. The observable signals come later: whether the next major model ships with a system card on time and undiluted, and whether the transparency role is refilled. If the disclosures get thinner each time, this essay will count for more than one former employee's frustration.
