Eh, well, stop building AI.

We're not going to do that, so it's pretty much unavoidable.

If I had to make a "most probable guess" of what the future looks like, we'll have some kind of AI mess caused that will cost billion of dollars. This will get governments up in a tizzy about their dangers, while still being hopelessly addicted to them in military and tracking their citizens. Most people think that governments will attack the models being built, but I don't think so, they want powerful models mostly for themselves. So they'll go after the next best thing. Hardware. What exactly that looks like, I'm not sure, but the big clusters of the machines will have to be registered and licensed. Likely the models will have to be encoded to run on particular hardware and getting caught with an unencoded model would be considered an illegal munition, like we did with cryptography in the past.