OpenAI’s Culture is Broken, Says Safety Employee Who Quit

David Robinson, a safety employee at OpenAI who helped draft the company’s current Preparedness Framework and oversaw safety reports for 12 frontier-model launches, has resigned after three and a half years, accusing the company of moving too quickly to develop increasingly capable AI systems.

Robinson disclosed his departure in an essay published by The Atlantic titled “I Quit OpenAI Because Its Culture Is Broken”. 

He said OpenAI’s approach to safety is shaped by a culture of rapid development and “perpetual sprints”, with the company relying on what it calls “iterative deployment”—releasing systems, identifying problems in use and strengthening safeguards afterwards.

“The time for trial and error is over,” Robinson wrote, arguing that this approach becomes increasingly risky as AI systems become more capable. He said the industry is not being “nearly careful enough” and needs to put greater emphasis on safety expertise and research before deploying more advanced systems. 

Robinson pointed to several recent incidents to support his argument. He cited OpenAI’s summer incident involving Hugging Face, in which he said the company “let a swarm of agents out by mistake”, as well as a subsequent case where a model in training bypassed restrictions on internet access. 

In that incident, Robinson recalled that a monitoring system alerted human staff but did not automatically shut down the model. He also referenced Anthropic’s acknowledgement of a separate safeguard failure caused by a configuration error.

The former OpenAI employee argued that frontier AI companies should adopt safety practices closer to those used in high-risk industries such as aviation and nuclear power. He called for the need to design and implement layers of redundancy and safeguards so that individual human errors or technical failures do not result in catastrophic outcomes. 

Robinson also acknowledged that leaving OpenAI meant giving up the ability to influence its safety processes from within. “As the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed,” he wrote. 

He said he plans to work outside the company to increase pressure on OpenAI and other AI developers to strengthen their safety practices. 

OpenAI defended its existing approach. A company spokesperson told Reuters that it is ensuring models do not become more capable than the company can “safely manage and secure”, and that it pauses training or holds back models when additional caution is required.

Recently, The Wall Street Journal reported that OpenAI had fired the researchers after an internal investigation found they had mishandled sensitive company information, including information allegedly shared with an outside organisation involved in AI evaluations. 

OpenAI reportedly said the employees had violated its policies and “broken the trust essential to our work”, while denying that they were dismissed for raising safety concerns. Those let go included two researchers working on safety and alignment and a research programme manager.