Geoffrey Hinton urges halt to AI over extinction level risks

Hinton’s latest interview matters less for its specific probability estimate and more because of who is talking and when. One of the architects of deep learning is now saying plainly, in a mass market European newspaper, that pausing AI development may be the only sane option and that current corporate stewards cannot be trusted to choose restraint. He links his warnings to fresh events, including Jacob Coxon’s resignation from Anthropic and internal alarm at both Anthropic and OpenAI about self improving systems. That moves talk of existential risk out of niche forums and into mainstream political discourse in the EU.

For the race to AGI, this raises the political cost of unbounded scaling. European regulators already view frontier labs with suspicion; Hinton’s comments give them intellectual cover to argue that slowing is not just precautionary but required. At the same time, he acknowledges AI’s upside for medicine, climate and productivity, which makes a blanket ban harder to sell. The likely outcome is more pressure for binding international coordination on model training thresholds and compute growth. In practice, that could channel capital away from unconstrained scaling races and toward labs that can credibly show they are prepared to hit a pause button if systems start to exhibit dangerous autonomy.