OpenAI safety employee resigns, calls company culture ‘broken’

David Robinson, an OpenAI employee who helped produce safety reports for the company’s major model releases, has resigned after three and a half years, arguing that OpenAI’s culture and the broader technology industry’s approach to AI safety are “broken.”

In an essay published by The Atlantic, Robinson said frontier AI developers rely too heavily on trial and error, or what OpenAI calls “iterative deployment.” He argued that this approach guarantees periodic failures whose scale will grow as AI systems become more capable.

Robinson called for frontier laboratories to operate more like nuclear power plants or busy airports, with redundant safeguards and slower, more deliberate planning designed to prevent human mistakes from causing disasters. He said reforms should extend beyond individual regulations and address a Silicon Valley culture characterized by rapid development, extreme confidence and optimism about increasingly powerful models.

He cited incidents involving AI agents, including a reported breach of Hugging Face systems, as evidence that current safeguards are inadequate. Robinson also said he had not encountered colleagues at OpenAI with professional experience in safety-critical fields such as aviation or nuclear energy.

OpenAI spokesperson Drew Pusateri said the company continues to strengthen its safety practices. Pusateri said OpenAI seeks to ensure its models do not become more capable than it can safely manage and secure, and that the company pauses training or withholds models when necessary.