Godfather of AI wants AI products to prove safety before launch

Geoffrey Hinton, the Nobel laureate known as the "Godfather of AI," says AI companies should prove their products are safe to regulators before launch, just like new drugs must pass FDA approval.

On a recent podcast, Hinton noted that drugmakers spend about a billion dollars to convince regulators before selling a new medicine, and AI should face the same bar. His comments came after OpenAI shelved its latest model, GPT-6.1 Astra, over worries it might overstep or ignore instructions.

OpenAI's safety lead said the model handles tough tasks better but still struggles with permission control and behavior logs. Hinton warned that AI is starting to improve itself in a loop, and exponential growth will amplify risks.

He guesses things could get much worse in just one or two years.