Geoffrey Hinton Proposes FDA-Style Safety Review Before AI Model Releases

Geoffrey Hinton, Nobel laureate and computer scientist, suggested on October 6 that AI companies should be required to prove their products are safe to regulators before release, drawing an analogy to the U.S. Food and Drug Administration's approval process for new drugs.

During an appearance on the podcast "Smart Girl Dumb Questions," Hinton noted that developing a new drug and bringing it to market involves extensive testing and roughly billion-dollar investments. He argued that AI should be held to a similar standard.

The proposal comes amid growing concerns over AI risks and OpenAI's decision to shelve its latest model after testing. Saachi Jain, OpenAI's head of safety systems, said in an earlier interview that GPT-6.1 Astra showed improvements in handling difficult tasks over extended periods, but still fell short in permission controls and behavioral log feedback.

Hinton also warned that AI is increasingly being used to improve itself, a recursive process that could accelerate capabilities while amplifying potential dangers. He estimated that the situation could become significantly worse within one to two years, though he acknowledged this was his personal speculation.