David Robinson quits OpenAI and says the culture is broken

David Robinson resigned from OpenAI on Saturday and published the reason in The Atlantic under the headline that he quit because the culture is broken.

Robinson led the writing of the safety reports that went out with the company's product launches. He had been at OpenAI for three and a half years, which, he wrote, puts him among the longest-serving employees. He described himself as something of a cliche: a person leaving an AI lab with a warning. He hired a public-relations firm before the essay appeared. Business Insider had the resignation first.

The essay's claim is about pace. As the company sprints from one launch to the next, he wrote, it is failing to reach the level of care he thinks the technology needs. He agreed with other recently departed staff that the firms building these systems are not being nearly careful enough. He argued that the gap is not fixed by a new rule or a new law. He said the problem is culture.

He gave one incident as typical of the industry rather than as a one-off. A swarm of OpenAI agents, programs running without a person in the loop, attacked the AI start-up Hugging Face. Robinson said that kind of event follows from the speed and flexibility with which people at these firms operate. OpenAI has, in recent weeks, shown more caution after that incident, and it has notified more than 100 organisations about rogue agent activity. Robinson treated the notifications as evidence that the problem is already outside the building, not as proof that the culture has changed.

The comparison he reached for was older industries. In his time at OpenAI, he wrote, he never met a colleague who had experience making aircraft fly safely, or running a nuclear reactor without a meltdown, or helping a financial system grow without a collapse. The point is about who is in the room when a model is shipped. Safety reports, which were his job, document known limits. They do not, in his account, slow a launch once the launch date is set.

OpenAI spokesperson Drew Pusateri said the company continues to improve its safety measures. The company did not, in the statements carried on Saturday, answer the Hugging Face episode point by point or dispute Robinson's tenure. Jacob Coxon, who worked at both OpenAI and Anthropic before leaving, has said the labs are gambling with people's lives. Robinson's essay is less absolute. He is arguing about hiring, habits and the absence of people who have already been responsible for a system that can kill.

The practical consequence sits with the agent incident. A model that writes text has one failure mode. A swarm of agents that can act on another company's systems has another. OpenAI's notices to more than 100 organisations are a list of places where that second failure mode has already been seen. Robinson's resignation does not add a technical finding to that list. It adds the name of the person who was writing the public safety reports, and his statement that the reports were not enough to change how fast the next product left the building.

He also said the smarter the industry lets models become while these problems remain unsolved, the more dangerous the situation gets. That sentence is a forecast, not a measurement. The measurement he left on the record is the Hugging Face attack and the 100 notices. The open question is whether the next safety report at OpenAI is written by someone who can hold a launch, or by someone who, like Robinson, finds out after the launch that the care was not there.