OpenAI safety researcher quits, urges nuclear-level AI oversight

A former OpenAI safety employee has resigned and is calling for nuclear-level safeguards for artificial intelligence, warning that the industry's rapid development approach is too risky. David Robinson criticized the company's reliance on trial-and-error methods in an essay published Saturday. Robinson, who spent three and a half years at OpenAI and helped draft its preparedness framework, argued that releasing AI systems and strengthening safeguards as problems emerge becomes increasingly dangerous as models grow more capable. He said advanced AI needs oversight comparable to nuclear power and aviation industries, and that capabilities are advancing faster than researchers' understanding of alignment. OpenAI disputed Robinson's characterization, stating it pauses training or holds back models when needed and is expanding third-party evaluations and real-time monitoring. The resignation follows OpenAI's disbandment of its Preparedness team in August, though the company said the work was redistributed, and comes amid growing scrutiny of AI testing and control practices.