The AI Safety Menu

All laws

U.S. superintelligence ban

Ban on superintelligence. Ban AI that outperforms people across most areas, and pause advanced development until regulators set rules.

Cuts lead by~17 monthsof America's lead over China, over 3 years
Cuts p(doom) by~0.6%from a 5% starting estimate
Holds up● Training● Lab's own use● Public release
Enact it on the menuSee the findings

What it does

No one may build AI that outperforms people across most areas. Under the Sanders–Casar bill, advanced AI development would also pause until a new federal Department of AI writes safety rules.

Applies to: Frontier AI development above compute thresholds set each year.

Where things stand

No lab has stopped frontier development. Polls show broad public support for going slower: a 2025 Future of Life Institute poll found 64% of Americans say superhuman AI should not be built until proven safe, or never.

Why it costs ~17 months

While the pause holds, the U.S. frontier barely moves. A year in, America is about 5.5 months behind where it would have been, assuming China does not join.

Biggest unknown: How long the pause would last and how much research would continue during it.

Why it lowers p(doom) by ~0.6%

Stops U.S. labs from building AI smarter than humans until there's a way to control it.

It directly stops U.S. labs from building the systems people worry about most. China isn't bound, and a ban is hard to define and enforce, so the range is very wide.

The strongest case that it costs more

A pause with no end date may last far longer than a year, and a permanent ban on the most capable systems would end the U.S. frontier altogether. If China does not join, the lead would be gone within the first year.

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.