OpenAI Safety Researcher Exits, Warns Industry Lacks Humility on AI Risk

David Robinson, the researcher responsible for penning OpenAI's safety assessments for each significant model release, has stepped down from his role. In a piece published in The Atlantic, Robinson is now raising alarms about what he views as a systemic breakdown in how the AI industry operates.

A Cultural Problem Runs Deep#

Robinson's critique goes beyond surface-level fixes. Rather than suggesting that new rules or light regulations might solve the problem, he argues the underlying issue is far more fundamental: the entire culture that dominates AI development is flawed at its core. The technology sector's characteristic combination of overconfidence and relentless speed—what he calls "extreme confidence" and "perpetual sprints"—pushes companies to build ever more powerful models under an assumption of unchecked progress. This worldview, Robinson suggests, glosses over or minimizes genuine risks.

Importing Safety from Outside Tech#

Robinson calls for AI firms to look beyond their own insular ecosystems. The move-fast mentality that defined much of Silicon Valley's success may not translate to building increasingly powerful AI systems. Instead, he argues, the industry should absorb lessons and structures from fields with proven track records in managing existential hazards.

Nuclear-Level Safeguards Are Needed#

Robinson specifically invokes a comparison to nuclear power plants and major airports as the model AI labs should emulate. These operations function with multiple layers of backup systems, deliberate planning processes, and built-in tolerance for human error—all designed to prevent individual mistakes from cascading into catastrophe. "Given today's risks, frontier labs need to run like nuclear power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster," Robinson states.

Part of a Broader Exodus#

Robinson is one of several prominent voices who have recently departed major AI organizations to speak publicly about their concerns. Jacob Coxon left Anthropic and warned AI could pose an existential threat. He was followed by Robert O'Callahan, Bilal Chughtai, Josh Engels at Google DeepMind, and Joe Benton at Anthropic, all raising similar alarms about the pace and direction of AI development.