David Robinson, who authored safety reports for OpenAI’s major model launches, announced his resignation this week. He detailed his concerns in an editorial published by The Atlantic, signalling a public break from the company. Robinson had been responsible for the safety documentation that accompanied each new release, a function that the firm highlighted as part of its risk-mitigation strategy. His departure marks the first high-profile safety staff exit in months.
Robinson characterises the prevailing culture in AI development as fundamentally flawed, arguing that superficial policy fixes cannot address the root problems. He says the sector operates with “extreme confidence” and “perpetual sprints,” pursuing ever larger models under unimpeded optimism that downplays or ignores emerging hazards.
He calls for frontier AI laboratories to adopt operational standards comparable to those governing nuclear power plants or busy airports. Such facilities would require multiple layers of redundancy, deliberate pacing, and exhaustive planning to ensure that inevitable human mistakes cannot open pathways to catastrophic outcomes.
Robinson’s exit follows a series of recent resignations among AI safety personnel at leading firms. Jacob Coxon left Anthropic and warned that AI “could kill us all by the end of the decade.” Subsequent departures included Robert O’Callahan, Bilal Chughtai and Josh Engels from Google DeepMind, as well as Joe Benton from Anthropic.
The Federal Trade Commission has reportedly opened investigations into both OpenAI and Anthropic, reflecting growing governmental scrutiny of AI development practices. Combined with internal warnings from former safety staff, these actions underscore mounting concerns that current industry dynamics may be insufficient to prevent high-impact risks.