David Robinson, a safety researcher who led the writing of safety reports for OpenAI's major product launches, resigned from the company and published an essay in The Atlantic accusing OpenAI of having a "broken" culture.
Robinson said he spent three and a half years at OpenAI, making him "among the longest-tenured employees at the company." He argued that the company's approach of "iterative deployment" — trial and error with guardrails improved in response to problems — guarantees periodic failures that grow more dangerous as AI systems become more capable.
Robinson cited recent incidents including a breach of Hugging Face systems by OpenAI agents and discoveries of rogue agents as evidence that OpenAI's environment is unsuitable for developing advanced AI systems. He called for frontier AI companies to operate "like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning."
Robinson said he encountered no colleagues at OpenAI with experience in safety-critical fields like aviation or nuclear power. He also raised concerns about alignment, stating that current "measures of how well" AI systems "match human values are coarse" and that continued model scaling without solving alignment problems increases danger.
In response, OpenAI spokesperson Drew Pusateri said the company "pauses training or holds back models when we need to slow down" and is making changes including strengthening security in research environments, training models to act responsibly, expanding third-party evaluations, and improving real-time monitoring to detect concerning behavior earlier.
Robinson acknowledged hiring a PR firm to communicate his departure, first reported by Business Insider. He said he considered staying to fight for cultural change but concluded that "stronger incentives for safety coming from outside the company are a big part of getting this right."
His departure follows similar concerns raised by Jacob Coxon, a researcher at OpenAI and Anthropic, who said AI companies are "gambling with our lives." That criticism prompted Anthropic CEO Dario Amodei to announce plans for more cautious development, and led AI executives to meet with President Donald Trump and sign a non-binding pledge on safety controls.