OpenAI Safety Architect Resigns, Warning of Broken Industry Culture
AI-generated image · Bay Street Wire
Former employee David Robinson calls for nuclear-level safeguards as a wave of safety researchers exit top AI labs.
David Robinson, who previously authored safety reports for every major model release at OpenAI, has resigned from the company and is publicly criticizing the industry's internal culture. In an editorial for The Atlantic, Robinson describes the current state of the AI sector as "fundamentally broken."
As The Verge first reported, Robinson argues that Silicon Valley has relied on "perpetual sprints" and "extreme confidence" to build increasingly larger models. He claims this "unimpeded optimism" has led the industry to ignore or underestimate potential risks. To combat this, Robinson suggests that frontier labs should operate with the same rigor as busy airports or nuclear power plants, implementing layers of redundancy and meticulous planning to ensure human errors do not lead to catastrophe.
Robinson's departure is part of a broader trend of safety personnel leaving prominent AI firms. The Verge notes that Jacob Coxon initiated this exodus by resigning from Anthropic and warning that AI could lead to total human extinction by the end of the decade. Other notable departures include Joe Benton from Anthropic, as well as Josh Engels, Bilal Chughtai, and Robert O’Callahan from Google DeepMind.