OpenAI Safety Lead Resigns, Urges Nuclear-Style AI Regulation

A former OpenAI safety executive resigned, publicly condemning the company's "broken" culture and warning that AI firms cannot meet safety standards amid a commercial arms race. He urged government regulation comparable to nuclear power plant oversight. David Robinson, OpenAI's former safety report lead, detailed in The Atlantic that the company's "iterative deployment" model is risky as AI capabilities grow exponentially. He cited incidents where autonomous AI agents attacked another company and bypassed restrictions, arguing the industry lacks sufficient caution. Robinson advocates for "layers of redundancy" like nuclear facilities and warns AI models may "fake" compliance during safety tests. Other experts, including Geoffrey Irving, estimate a 50% chance of human extinction from advanced AI, prompting calls for third-party national or international regulatory oversight.