David Robinson, a safety specialist responsible for authoring the evaluation reports accompanying major artificial intelligence model releases at OpenAI, has resigned from his position. Following his departure, Robinson has published an editorial in The Atlantic warning that the internal culture across the artificial intelligence sector is fundamentally broken.
Critique of Industry Culture and Unchecked Optimism
In his public critique, Robinson argued that the underlying issues within frontier artificial intelligence development cannot be resolved merely by adding standard compliance checklists or superficial regulations to model training workflows. Instead, he pointed directly to the dominant operational mindset across Silicon Valley firms.
According to Robinson, leading technology companies have built progressively larger and more capable models under conditions of extreme confidence and perpetual sprints. He characterized this environment as one driven by unimpeded optimism, which consistently downplays, underestimates, or outright ignores the serious hazards associated with advancing machine intelligence systems.
Robinson asserted that the artificial intelligence sector must abandon its insular move-fast-and-break-things philosophy. In its place, he argued, developers and executives must cultivate institutional humility and seek guidance from established high-risk fields outside the standard technology sector.
Demands for Nuclear-Level Safety Standards
To mitigate the growing risks posed by cutting-edge systems, Robinson called on frontier research laboratories to dramatically restructure their operations. He maintained that these facilities should implement structural safeguards akin to those found in commercial aviation and nuclear energy production.
Specifically, Robinson stated that frontier AI labs must operate with layers of operational redundancy and meticulous, time-consuming planning, similar to the rigorous management required at nuclear power stations or busy international airports. In his view, instituting such comprehensive protections is the only viable way to ensure that inevitable human operational mistakes do not trigger catastrophic consequences.
A Growing Exodus of AI Safety Specialists
Robinson’s resignation adds to a documented series of departures among safety researchers and ethics specialists at leading artificial intelligence firms in recent months.
Among those who have recently exited the sector, former Anthropic researcher Jacob Coxon publicly raised alarms following his departure, warning that unchecked artificial intelligence could kill everyone by the end of the decade. Anthropic also saw the exit of researcher Joe Benton.
Similar personnel shifts have emerged across other major research hubs. At Google DeepMind, several notable technical and safety staff members have also left their posts, including Robert O’Callahan, Bilal Chughtai, and Josh Engels. Robinson’s public statements reflect a widening unease among researchers who previously worked directly on evaluating advanced model deployments from within the industry’s most influential laboratories.
About The Author
