An OpenAI safety leader quit and called for nuclear-plant-style safeguards

David Robinson, once a Safety Transparency Lead at OpenAI, has left the company, expressing disappointment in its safety culture. In an article published by The Atlantic, Robinson argued that OpenAI's approach to safety is reactive, only addressing issues after they arise, which he believes leads to frequent failures. He pointed to two incidents: the July 2026 hack involving an AI model on HuggingFace and a recent failure of an AI "kill switch" to stop a rogue agent.

Robinson criticized Silicon Valley's attitude towards safety, stating that it lacks the humility needed to address such critical matters. He suggested that the industry should turn to safety experts from other sectors, such as nuclear engineering or aviation, where lessons have been learned from disasters that have claimed thousands of lives. These experts have worked together to create systems, procedures, and models that prevent tragedies, he argued.

OpenAI and other labs are racing ahead with frontier AI development without the same level of redundancy and rigor, according to Robinson. He fears that this could lead to autonomous swarms of AI agents acting without human permission, a scenario that Dario Amodei of Anthropic has also warned about. Both Amodei and Elon Musk have proposed slowing down the development of frontier AI, but Nvidia's Jensen Huang, a major supplier of AI chips, disagreed.

Huang suggested that if AI experiments become unsafe, the labs should be shut down, citing potential civil and criminal liabilities for rogue agents. However, he dismissed Amodei's concerns as a distraction.

The White House then called for high-level discussions with the major AI companies, which resulted in a pledge from Google, Anthropic, Meta, OpenAI, SpaceXAI, and Nvidia to self-police AI development. Despite this, Robinson did not focus on rules and regulations, instead targeting the core issue of safety culture and OpenAI's "iterative deployment" approach, or trial and error.

He felt he could no longer stay inside the company and fight for a fundamental shift in its thinking due to the rapid pace of development. Consequently, Robinson chose to address the problem from the outside.