OpenAI: Departure of a Former Security Staff and Concerns About Culture

David Robinson, a former security report writer at OpenAI, has left the company and is criticizing a culture he deems flawed. He is calling for "nuclear-level" safety practices. Other specialists have also departed from major labs, as a reported FTC investigation is underway concerning OpenAI and Anthropic.

Reported Federal Investigation and Series of Departures from Major Labs

The FTC is said to have opened an investigation into OpenAI and Anthropic. Meanwhile, several security specialists have left major AI labs. At Anthropic, Jacob Coxon was one of the first to depart, publicly stating that AI "could kill us all by the end of the decade." Robert O’Callahan, Bilal Chughtai, and Josh Engels have left Google DeepMind, while Joe Benton has also exited Anthropic. David Robinson has resigned from OpenAI.

Robinson's Diagnosis: A "Broken" Culture and Optimism that Overlooks Risks

David Robinson describes OpenAI's culture as "broken" and believes the industry has fundamental failings. He argues that the issue goes beyond simply adding new rules for model training. According to him, Silicon Valley has operated with "extreme trust" and "perpetual sprints," developing ever-larger models with "unfettered optimism" that minimizes or ignores potential problems. He finds the cynicism towards those who raise alarms after having contributed to these technologies understandable, but believes these warnings should not be ignored.

Required Approach According to Robinson: Humility and "Nuclear-Level" Safety

David Robinson recommends applying safety standards comparable to those in the nuclear industry to security systems. He argues that leading labs should adopt the operational modes of nuclear power plants or high-traffic airports, layering numerous redundancy measures and organizing rigorous, long-term planning to prevent an inevitable human error from causing a catastrophe. He also urges AI companies to demonstrate humility and to extricate themselves from the closed and fast-paced environment of the tech industry. He presented these arguments in an editorial published in The Atlantic.