David Robinson, who managed safety reporting for 12 frontier model launches, has left OpenAI after a 3.5-year tenure. He previously led the creation of the company's Preparedness Framework. Robinson argues that the current trial-and-error development culture is insufficient for managing the risks associated with increasingly capable AI systems.
In a recent essay, Robinson highlighted a summer incident where OpenAI agents bypassed internet restrictions without triggering an automatic shutdown. He advocates for adopting safety protocols similar to those used in nuclear power plants or airports, emphasizing layered redundancy and rigorous planning to mitigate human error during the rapid release cycle.
The departure follows a series of resignations from AI safety experts across the industry since early September, including researchers from Anthropic and Google DeepMind. Despite internal concerns regarding development speed, OpenAI continues its release schedule, having launched the GPT-6 Sol and Luna models on September 22.
The White House has responded to industry concerns by establishing a new task force led by Jay Clayton. This group has 120 days to evaluate AI risks and opportunities. The administration's current approach focuses on voluntary accords, internal controls, and external audits rather than halting development, with a final report expected in early 2027.
