The AI Safety Menu

All laws

U.S.–China safety deal

U.S.–China AI safety agreement. Washington and Beijing agree on AI red lines, share safety testing and set up a crisis hotline.

Lead costAlmost noneof America's lead over China, over 3 years
Cuts p(doom) by~0.1%from a 5% starting estimate
Holds up○ Training○ Lab's own use○ Public release
Enact it on the menuSee the findings

What it does

The two governments would agree on a short list of things no AI should be allowed to do, such as helping build bioweapons or copying itself without permission, run some joint safety tests, and set up a hotline to warn each other about dangerous AI incidents. It would not stop either country from building more powerful AI.

Applies to: Government-to-government commitments.

Where things stand

There is no hotline, no joint testing and no agreed red lines between governments. Unofficial scientist dialogues (IDAIS: Beijing 2024, Venice 2024, Shanghai 2025, London April 2026) have produced consensus statements only.

Why it costs almost no lead

Talks and a hotline cost labs almost nothing. Under a day.

Biggest unknown: Whether any deal would contain verifiable commitments rather than just a communication channel.

Why it lowers p(doom) by ~0.1%

Sets shared red lines, joint safety testing and a crisis hotline between the two AI superpowers.

Shared red lines and a hotline lower the chance of an accident spiralling. Without verification, the effect is modest.

The strongest case that it costs more

China may use talks to slow U.S. policy without changing its own behaviour. Critics worry about legitimizing Chinese AI governance.

The debate

For

Against

Sources

Rough starting points, not precise forecasts. Lead costs assume China doesn't depend on U.S. models, the case least favorable to safety laws, and count 3 years. On the menu you can change every assumption and put in your own numbers. Last priced 2026-09-26.