The AI Safety Menu

All laws

Cut government AI testing

Cut government AI safety testing. Cut the budget and staff of the government's AI testing center.

Lead costAlmost noneof America's lead over China, over 3 years
Raises 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

Government would do less or narrower testing of frontier models for dangerous abilities, relying more on companies' own checks. Money and staff for independent evaluators would shrink or stay flat.

Applies to: Federal AI testing programs.

Where things stand

CAISI runs on about $15M a year ($10M FY26 appropriation plus a $10M loan), with voluntary pre-release access under a June 2, 2026 executive order (up to 30 days, closed models only).

Why it costs almost no lead

Labs spend a little less time with government testers: under a day gained.

Biggest unknown: Whether CAISI's small, voluntary, closed-model-only testing is enough to catch dangerous capabilities before release, or whether it will be cut further in final FY27 appropriations.

Why it raises p(doom) by ~0.05%

Not a safety law. Shrinks the government office that tests frontier models for dangers.

Government testers are one of the few outside checks on frontier models before release.

The strongest case that it costs more

Supporters argue testing should focus on national security and that labs test more thoroughly than government can.

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.