Geoffrey Hinton says AI models should face safety checks before release, comparing the idea to the US Food and Drug Administration’s role in approving drugs. His account of a recent senators’ briefing also points to a tougher question: can public pressure move faster than legislation?
Geoffrey Hinton Watch analysis
What happened
In an interview on the podcast Smart Girl Dumb Questions, Hinton backed a proposal raised by physicist Max Tegmark: companies should have to demonstrate an AI system’s safety to a government body before releasing it. Hinton called that “the very least we should have for AI”, according to Yahoo Finance’s account of the interview.
Hinton said he left a recent closed-door briefing with 19 US senators “maybe a little bit more optimistic”, while noting that the group was largely self-selected. He argued that public understanding and pressure, rather than lawmakers acting alone, would be crucial to meaningful regulation. He compared that potential shift to the pressure that built around climate change.
Why it matters
The proposal would move oversight towards checking systems before they reach the public, rather than relying chiefly on companies to set their own rules or regulators to respond after release. The interview account does not spell out what a safety test would involve or how such a system would work in practice. Those details are where a promising analogy meets the less tidy business of writing rules.
Hinton also said he does not regret his research itself, but wishes he had recognised sooner how dangerous AI systems could become. He said he began warning publicly in 2023.
Our read
Pre-release checks are a serious policy proposal, not a magic safety stamp. Hinton’s endorsement gives the idea weight, but the hard questions are still about who sets the standard, what evidence companies must provide and whether checks can keep pace with changing systems. Readers should weigh the principle separately from the as-yet-unspecified machinery.
What to watch
- Whether lawmakers or regulators develop concrete proposals for pre-release AI safety checks.
- What kinds of evidence a company would have to provide, and who would assess it.
- Whether public pressure translates into policy, as Hinton argues it must.
Discussion spark: Should powerful AI models have to pass government safety checks before release, or would that give regulators too much control over a fast-moving technology?
Sources and evidence
Independent WittyWires tracker for public updates about Geoffrey Hinton. Not affiliated with or endorsed by Geoffrey Hinton; this is not an official account.