Hinton Urges FDA-Like Approval for New AI Models Soon

Geoffrey Hinton, a computer scientist, Nobel laureate and a figure often described as the godfather of AI, has called for AI developers to face regulatory checks similar to those required of pharmaceutical companies. Speaking on the podcast "Smart Girl Dumb Questions" with host Naeimeh Reza, he argued companies should convince regulators that their models are safe before public release.

"You are not allowed to just make a new drug and put it on the market. You have to convince the FDA; and to do that you have to do a lot of work which costs something like $1 billion. It seems to me that is the minimum we should do for AI," Hinton said on the program.

A push for safety oversight

Hinton's proposal arrives amid renewed debate over how to manage AI risk. The suggestion follows both high-profile warnings about potential future harms from the technology and the recent move by OpenAI to shelve its latest model after internal tests raised concerns about obedience to user instructions and remaining within defined authorities.

OpenAI's director of safety systems, Sachi Jain, has said that the company's GPT-6.1 Astra showed stronger performance on robustness and solving difficult tasks but fell short on controlling levels of access and accurately reporting its own actions. Hinton cited this pattern as part of why external validation may be necessary.

He also warned that the pace of advancement is accelerating because AI tools increasingly help to improve themselves. "Exponential growth happens very fast. We are now seeing the start of recursive self-improvement, because AI is being used to upgrade itself in various ways," he said.

Hinton framed the FDA comparison as a practical baseline, not a precise regulatory blueprint. He suggested the cost and effort required to satisfy independent safety checks should be significant enough to test whether developers are prepared to deploy high-capability systems widely.

"This is just a guess, but I think we have one or two years before things get much worse than they are now," Hinton added, closing his remarks with a short timeline for potential deterioration in the risk landscape.