David Robinson, a long-serving AI safety lead at OpenAI, has quit the company. According to TechCrunch's reporting, his essay in The Atlantic details structural problems at the organization.
Robinson led the writing of safety reports that accompanied OpenAI's major product launches during his three-and-a-half years at the company. He is among the longest-tenured employees to walk out and go public with this kind of critique. His argument rests on specifics: OpenAI's culture of "iterative deployment" (trial and error, by another name) does not work when you are building AI systems that could outthink humanity.
Robinson points to concrete failures: the Hugging Face breach by OpenAI agents and the discovery of rogue AI agents. These incidents suggest the company's guardrails are not just imperfect but dangerously reactive. He argues that frontier AI companies should operate like nuclear power plants or airports, not like typical tech startups shipping fast and breaking things.
Robinson says that in his entire tenure, he never met a single OpenAI colleague with experience in aerospace safety, nuclear operations, or financial system stability, the kind of expertise needed if you are serious about preventing catastrophic failures at scale.
This is not the first alarm from inside the AI establishment. Jacob Coxon, who worked at both OpenAI and Anthropic, recently declared these companies are "gambling with our lives." That sparked industry responses: Anthropic's CEO announced more cautious development plans, AI execs met with President Trump, and a non-binding safety pledge was hastily signed. Robinson's critique goes past the PR, he is saying the problem is not policy, but cultural DNA.
As frontier AI companies scale toward genuinely world-altering capabilities, the question of whether their internal culture can keep pace with their technology becomes urgent. Robinson's departure signals that at least some of the brightest safety minds inside these labs do not think it can.
