How Middle Powers Like Poland Could Police AI Chips

A recent course on compute verification has sparked a debate about who should enforce rules on AI chips. The idea is to create a global system that tracks and limits how much computing power is used for AI training, to prevent misuse. But who should run it? A new proposal argues that middle-power countries—like Poland—could be the perfect referees.

Middle powers are countries without big AI companies or chip factories, but they still have a stake in keeping AI safe. The proposal suggests they form a coalition to issue licenses for AI chips. These licenses would limit how much computing a chip can do and where it can operate. Since middle powers have nothing to gain from cheating, they would be trusted to enforce the rules fairly.

Another key role for middle powers is to act as independent auditors. The verification system would require installing monitoring hardware in AI facilities. But if the country supplying that hardware has its own agenda, it could spy on others. Middle powers, with no hidden interests, could ensure the hardware is used only for its intended purpose, building trust in the system.

The catch? This only works if major powers like the US and China agree to the rules and give the coalition real authority. Without their support, the coalition can't enforce anything. Still, the proposal offers a way to bring more countries into the AI safety effort and make the system more legitimate. For everyday people, this could mean a safer AI future with fewer risks of misuse.