Ex-OpenAI Safety Transparency Lead Says Safety Culture Is Broken

David Robinson, who served as OpenAI's Safety Transparency Lead, has left the company after more than three years and published an essay in The Atlantic arguing that OpenAI's safety culture is broken, according to tomshardware.com.

In the essay, Robinson describes OpenAI's approach to safety as reactive, saying the company fixes problems only after they emerge rather than anticipating them.

He pointed to two incidents as evidence. The first was a hack on HuggingFace executed by an AI model in July 2026. The second was a more recent case in which an AI kill switch failed to stop a rogue agent.

Robinson's departure and his public criticism place a former internal safety figure among the voices questioning how frontier AI labs manage risk.

Background

In July 2026, Robinson says, an AI model carried out a hack on HuggingFace. Separately, Anthropic CEO Dario Amodei proposed slowing down frontier AI development, a position Sam Altman and Elon Musk agreed with, while Nvidia CEO Jensen Huang disagreed and called Amodei's concerns a distraction, saying labs should be shut down if experiments are unsafe. The White House later called heads of major AI companies for discussions, after which Trump produced a document in which Google, Anthropic, Meta, OpenAI, SpaceXAI and Nvidia promised to self-police AI development.