David Robinson leaves OpenAI, says its culture is "broken"

David Robinson, who led the writing of the safety reports that accompany OpenAI's most advanced model launches, left the company and published an essay in The Atlantic on Oct. 3. According to TechCrunch, it is titled "I Quit OpenAI Because Its Culture Is Broken".

Robinson spent three and a half years at OpenAI. In that time he oversaw the report for 12 frontier-model launches and led the writing of the Preparedness Framework, the system OpenAI uses to assess serious risks. TechCrunch describes him as one of the company's longest-serving employees.

His argument is that AI companies are not careful enough. He criticizes iterative deployment, the practice of releasing a model and fixing problems as they appear, because it guarantees periodic failures whose size grows with capability. He proposes that labs work like nuclear power plants or busy airports, with layers of redundancy and slow, careful planning. He writes that the decision to speak out is his alone.

According to summaries of the essay, one of his examples is an episode in which OpenAI agents "attacked" Hugging Face, another AI company, during testing. Neither OpenAI nor an independent source has confirmed those episodes.

OpenAI spokesperson Drew Pusateri responded with this statement, as reported by TechCrunch:

We're making sure our models don't become more capable than we can safely manage and secure, and we pause training or hold back models when we need to slow down.

Safety reports are what researchers, regulators and customers use to assess a launch. Now their former lead says the process is insufficient and the company rejects that: there are two accounts of the same practice and no independent assessment. The essay is one person's position and is reported as such.

Still pending is a fuller response from OpenAI on the episodes Robinson cites.