Geoffrey Irving calls for stopping AI progress now, not at the last minute

On 22 October, researcher Geoffrey Irving argued on X that AI progress should be stopped now. He proposes starting with the key companies, moving on to more US companies, and reaching China through international agreements.

Irving makes clear he is not saying this because today's models could kill everyone, but because those coordination steps take time. Acting now leaves more of it for the later ones.

By his account, the usual reply from people at AI companies is to wait until the last minute, when the very next model is dangerous. Irving answers that this assumes governments can coordinate internationally on a dime, which clashes with what those same people think of governments. He says he spent two and a half years in the UK government.

This is one researcher's opinion, not a measurement or an official policy.