A former safety employee at OpenAI has publicly criticised the company's approach to AI safety, arguing that a fast-paced culture focused on rapid development raises the risk of serious failures. David Robinson, who recently resigned, made the case in an essay titled "I Quit OpenAI Because Its Culture Is Broken," published by The Atlantic on Saturday.

Robinson wrote that AI companies, including OpenAI, are not being "nearly careful enough" and should place greater emphasis on safety expertise and research before building more capable systems. "The time for trial and error is over," he wrote, arguing that advanced AI systems call for safeguards closer to those used in industries such as nuclear power and aviation. He said OpenAI relies heavily on what it calls "iterative deployment": releasing systems to the public and strengthening safeguards when problems emerge. "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.

Robinson says he spent three and a half years at the San Francisco-based company, where he helped draft its preparedness framework, an internal policy for assessing risks from its most advanced models. He also oversaw safety reports for 12 launches of so-called frontier models, the most capable systems at the cutting edge of the technology. He further warned that AI capabilities are advancing faster than researchers' understanding of alignment, the field concerned with ensuring that AI systems act in line with human goals and values.

OpenAI responded with a statement from a spokesperson: "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."

The resignation adds to a continuing debate within the AI industry over whether companies are moving too quickly to develop increasingly powerful systems. OpenAI and its rival Anthropic have both faced scrutiny after incidents in which safety controls failed or experimental systems behaved unexpectedly.

Why it matters: the dispute pits two views against each other. One holds that releasing systems step by step and fixing problems as they surface is a practical way to learn. The other, which Robinson now advances, holds that the stakes with highly capable systems are too high to learn from mistakes after the fact. Public criticism from someone who helped shape a leading developer's safety processes is likely to sharpen that discussion.