David Robinson, a former safety employee at OpenAI, has resigned from the company and publicly criticized its approach to AI development. In an essay published by The Atlantic, Robinson said OpenAI’s culture is “broken” and argued that its reliance on trial and error is becoming too risky as AI systems grow more capable. His comments add to growing debate around safety inside major AI companies.
David Robinson says OpenAI relies too much on trial and error
Robinson spent three and a half years at OpenAI and said he helped lead the writing of safety reports that accompanied major product launches. He described himself as one of the company’s longest-serving employees before deciding to leave.
In his essay, Robinson criticized what OpenAI calls “iterative deployment,” where AI systems are released, problems are identified, and safeguards are improved in response. He argued that this approach naturally allows failures to happen, while the potential impact of those failures grows as AI models become more capable.
Don’t miss the best of The Mac Observer
Set us as a preferred source and our Apple reporting ranks higher in your Google Search results and Discover feed — one tap, no account changes.
Robinson also pointed to recent incidents involving OpenAI agents, including a breach of Hugging Face systems, as examples of why he believes the industry needs to be more cautious. He argued that companies developing advanced AI should operate more like nuclear power plants or busy airports, with multiple layers of safety and careful planning.
His concerns extend beyond OpenAI. Robinson said the wider AI industry has placed too much value on optimism and speed while not giving enough attention to whether increasingly capable systems will remain aligned with human goals. The issue is also part of the wider discussion around AI safety and development.
OpenAI says it is strengthening safety measures
OpenAI spokesperson Drew Pusateri said the company continues to improve its safety work. He said OpenAI pauses training or holds back models when necessary and is making changes to security in its research and testing environments.
The company also said it is training models to complete tasks responsibly, expanding work with third-party evaluators, and improving real-time monitoring to identify concerning behavior earlier in the training process.
Robinson, however, believes technical changes alone will not solve the problem. He argued that the culture around AI development also needs to change, with stronger incentives for safety coming from outside companies.
His resignation adds another public voice to the debate over how quickly companies such as OpenAI should develop increasingly powerful AI systems. Robinson said he considered staying to push for changes from inside the company, but felt that his team was too focused on moving quickly to make broader changes.
The dispute ultimately reflects a larger question facing the artificial intelligence industry: whether current safety practices can keep pace with increasingly capable AI systems.
Discussion