World Leaders Move to Ban Superhuman AI — Here's the Catch

Politicians are starting to treat superhuman AI — machines smarter than all of humanity combined — as something to outlaw rather than race toward. In the UK, a bill that would ban building superintelligence on British soil passed its first reading in the House of Commons on 8 September. In the US, Senator Bernie Sanders has called for a similar law. On 16 September, the UN Secretary-General and the European Commission President both urged countries to coordinate on AI safety.

The problem is that banning new AI training runs (the expensive process of building a model) isn't enough. Modern AI also gets smarter simply by being used — something called inference, which is what happens every time you type a prompt. Give an existing model enough computing power and it can improve without anyone building anything new. So a real ban would have to restrict how today's most powerful models are used, too. And defining exactly where "superintelligence" begins is, by the author's own admission, currently impossible.

A total worldwide ban creates its own trap. It would freeze the biggest AI companies: no new models, limits on using the ones they already have, and nowhere to relocate, since the rule would be global. Pressure from those firms, and from a public that likes cheap AI tools, would build until governments reopened the door.

The author's suggested fix is a release valve: allow a small, carefully chosen set of AI projects that outside inspectors can verify. The real question, then, isn't whether to ban superintelligence — it's how to let some AI development continue without losing control of it.