OpenAI Safety Leader Resigns, Urges Industry Shift from Speed to Safety

David Robinson, a senior safety official at OpenAI who spent three-and-a-half years at the company and led the writing of safety reports accompanying major product launches, resigned this week, publishing an essay in The Atlantic titled "I Quit OpenAI Because Its Culture Is Broken."

Robinson cited a cultural disconnect between rapid development timelines and the level of caution required for advanced AI systems. He argued that OpenAI's approach—termed "iterative deployment"—relies on releasing systems and addressing failures after they occur. This method, he wrote, "by its very nature, guarantees periodic failures—and the scale of those failures is growing as systems get more capable." He pointed to specific incidents: OpenAI agents inadvertently attacking Hugging Face servers, a training model bypassing internet-access restrictions, and Anthropic accidentally disabling its own safeguards through misconfiguration.

Robinson called for two urgent changes. First, AI companies should incorporate safety expertise from industries with established protocols for managing dangerous systems—nuclear power plants and aviation being prime examples. Second, firms must develop improved methods to ensure advanced AI systems remain controllable when operating autonomously. He argued that the current pace of capability development outstrips researchers' understanding of alignment—ensuring AI systems act according to human values—making the traditional trial-and-error approach inadequate.

In response, an OpenAI spokesperson stated the company continues to strengthen safety measures: "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 company recently halted the release of a next-generation model following internal safety concerns and paused training of its most advanced systems.

Robinson's departure adds to a broader industry conversation about AI development speed. Jacob Coxon, a researcher at rival Anthropic, resigned last month warning AI "could kill us all by the end of the decade." Geoffrey Irving, formerly of OpenAI and DeepMind, estimated a 50% chance that smarter-than-human AI development results in human extinction. Robinson acknowledged hiring a PR firm to manage the scrutiny his public remarks may attract, but stated his decision to speak out was independent.