An OpenAI safety employee has quit and is sounding the alarm

David Robinson, who until recently prepared the safety reports released alongside OpenAI's major model updates, resigned this week and used an Atlantic editorial to outline a stark critique of the AI sector. Robinson told readers that the industry's prevailing mindset—marked by extreme confidence and relentless rapid development—has produced a culture that systematically underestimates potential harms. He framed the problem as deeper than technical fixes or modest regulatory changes. Robinson urged AI developers to adopt greater humility and to broaden their perspective beyond the insular, move-fast ethos common in Silicon Valley. He warned that simply adding new rules for training models will not be enough and called for operational standards modeled on institutions that manage high-consequence risk. Specifically, Robinson said frontier AI labs should run like nuclear power plants or busy airports, with multiple redundant safeguards and deliberate, time-consuming planning to reduce the chances that human error leads to disaster. His comments follow a recent wave of departures by researchers and safety staff from leading AI firms. The Verge notes earlier exits that include Jacob Coxon from Anthropic, who publicly warned of existential risks, and departures from Google DeepMind (Robert O’Callahan, Bilal Chughtai, Josh Engels) and Anthropic (Joe Benton). Robinson’s move adds to that pattern and amplifies internal concerns about how major labs balance rapid capability development with risk mitigation. The critique from a former internal safety lead highlights ongoing friction between fast-paced product development and calls for stricter governance and operational discipline. Whether industry practices will shift in response remains uncertain, but Robinson’s public resignation and editorial contribute to a growing chorus of insiders urging systemic change in how frontier AI is managed.