On Oct. 6, Geoffrey Hinton proposed that AI companies persuade a regulator their products are safe before releasing them. Speaking on the Smart Girl Dumb Questions podcast, he compared the idea with drug approval—a review that would happen before a product reaches the market.
What Geoffrey Hinton proposed
Hinton’s proposal would put a safety review before the release of AI products: companies would have to convince a regulator that their products are safe. It is a proposal, not a rule in force.
The FDA enters Hinton’s argument as part of the drug-approval comparison. His AI proposal calls for review by a regulator; the comparison does not name the FDA as that regulator.
The drug-approval comparison
Hinton compared the work involved in persuading the FDA to approve a new drug with the safety hurdle he proposed for AI. He put the drug-related work at about $1 billion. That figure belongs to his drug analogy.
Why Hinton raised the idea
Hinton said AI was being used in multiple ways to improve AI, a process he described as recursive self-improvement. He argued that this could accelerate both capabilities and risks.
He guessed that conditions might become much worse within a year or two, and described that timeline as a guess.
