The technology that could make an international AI slowdown possible

If the US and China agreed tomorrow to stop training frontier AI, how would either side know the other had actually stopped? Right now, says Tom Milton, the answer is crude: “I think we would probably just shut down the data centres.” That would also kill everything else those chips do, which is why any slowdown is so hard to agree to. Tom is CEO of Amodo Design, a UK hardware lab building tools to check what AI data centres are really doing. His pitch is that verification makes a slowdown affordable: if you can prove a cluster is only running existing models and not training new ones, most of