“OpenAI’s culture is broken”: David Robinson, who drafted the security reports for launches, has resigned

In an opinion piece published by *The Atlantic*, one of the company’s longest-serving employees calls for a level of rigour on a par with that found in the nuclear or aviation industries, rather than a trial-and-error approach.

David Robinson, who led the writing of the security reports accompanying OpenAI's major product launches, left the company after three and a half years, which ranked him among the longest-serving employees. He explains this in an article published on October 3 by The Atlantic, where he says that the company's culture is "broken".

Failures guaranteed by the method

He says OpenAI is moving forward through trial and error, a move that makes periodic failures inevitable. He cites the recent intrusion of OpenAI agents into Hugging Face's systems and the continued discovery of other uncontrolled agents within the company. State-of-the-art AI labs, he argues, should operate like nuclear power plants or large airports, with multiple layers of redundancy and careful planning. He says he's never met a colleague with experience in aviation security or operating a reactor.

Robinson acknowledges that he has surrounded himself with a PR agency, but says the decision to speak out is entirely up to him. OpenAI spokesman Drew Pusateri said the company is ensuring that its models don't become more capable than it can handle, citing security reinforcements, training adjustments, external evaluations and increased oversight. The resignation follows warnings from researcher Jacob Coxon that AI companies are "playing with our lives."