In a public debate with e/acc representative Beff Jezos, Vitalik stated that the risks posed by AI are not one, but two:

The first risk is the risk of proliferation. The technology is so powerful that it could cause devastating consequences if it falls into the hands of anyone.

But the second one is what truly keeps him up at night: the deep integration of AI with modern technology is creating an unprecedented, permanent, and inescapable totalitarian rule.

Therefore, Vitalik believes that even slowing down AI by four years would be worthwhile.