The AI Safety Menu

All laws

Yearly audit

Annual independent compliance audit. An outside auditor checks once a year that you did what your framework says.

Lead costAlmost noneof America's lead over China, over 3 years
Cuts p(doom) by~0.05%from a 5% starting estimate
Holds up○ Training○ Lab's own use○ Public release
Enact it on the menuSee the findings

What it does

Once a year, an outside auditor checks whether the lab actually did what its safety framework promised. A summary is made public.

Applies to: Covered frontier developers (SB 1047 would have started this in 2026).

Where things stand

Labs already commission external security audits and, increasingly, external reviews of their safety frameworks. Anthropic commits to seeking public external review of its Risk Reports.

Why it costs almost no lead

An audit looks back at the past year, so nothing waits on it. The effect is under a day.

Biggest unknown: Whether audit findings ever require remediation before a launch.

Why it lowers p(doom) by ~0.05%

Checks that labs actually follow their own safety promises.

Audits look back at whether labs kept their promises. That keeps promises honest but catches few dangers directly.

The strongest case that it costs more

An auditor who has to sign off on a framework may demand conservative interpretations of thresholds, which ratchets a lab's own gates tighter. Over several generations that ratchet could add up. It is a second-order effect and we have no data on it.

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.