The AI Safety Menu

All laws

Liability insurance

Mandatory liability insurance for AI developers. Frontier developers must insure against the harms their systems could cause.

Cuts lead by~5 daysof 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

Frontier AI developers must carry insurance against harms their systems could cause, and the insurer, not the developer, chooses who audits their safety practices.

Applies to: Frontier AI developers.

Where things stand

No U.S. bill requires it. Specialist insurers such as AIUC certify and insure AI agents voluntarily. Many general insurers are adding AI exclusions to their policies.

Why it costs ~5 days

Premiums cost money, not training time. But insurers charge more for riskier practices, so labs move some research effort into safety to keep premiums down: about 2 days of progress a year. The bigger risk is that insurers won't cover AI at all.

Biggest unknown: Whether insurers would offer coverage at all. Many are currently excluding AI risks from their policies.

Why it lowers p(doom) by ~0.05%

Makes insurers, who would pay for disasters, check that labs are careful.

Insurers have money on the line and would push for audits of everyday risks. But no insurer can pay out after a true catastrophe, so they can't price the risks that matter most for p(doom).

The strongest case that it costs more

Catastrophic AI losses could exceed what private insurance can cover. A mandate without available coverage becomes a de facto ban on deploying the most capable models.

The debate

For

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.