Liability for AI harms

Product liability for AI harms. Victims can sue AI companies when their systems cause harm, under federal product liability law.

Cuts lead by~9 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 menu See the findings

What it does

AI systems are treated as products under the law. If one causes harm, the people hurt can sue the company that built it, the same way they could sue over a faulty car part.

Applies to: Developers and deployers of AI systems.

Classify AI systems as products for liability purposes.
Create a federal cause of action for product liability claims when an AI system causes harm (the AI LEAD Act, S. 2937).

Where things stand

AI companies already face ordinary lawsuits, and many are pending. What they do not face is a clear federal rule that treats AI systems as products, which would make those suits easier to bring and win.

Why it costs ~9 days

Lawsuits don't stop training. But labs that pay for harms spend more of their research effort on testing and safety, and ship more cautiously. We estimate that care costs about 3 days of progress a year.

Biggest unknown: How much caution the threat of lawsuits actually buys, and whether it spills over from product decisions into research.

Why it lowers p(doom) by ~0.05%

Makes labs pay for serious harms their models cause, so they have a reason to prevent them.

Paying for harms gives labs a reason to be careful. But no company can pay for a true catastrophe, so the incentive is weakest where it matters most.

The strongest case that it costs more

Open-ended liability can freeze a whole field. If juries treat any misuse of a general-purpose model as a product defect, labs could slow every release, stop publishing open models, and move research abroad. That could cost months, not weeks.

The debate

For

Sens. Durbin and Hawley introduced the AI LEAD Act to let victims sue AI companies, 2025.
American Association for Justice, Encode AI and others endorsed the AI LEAD Act, 2025.
Gabriel Weil, Touro Law proposed strict liability for advanced AI as an abnormally dangerous activity, 2024.

Against

NetChoice called treating AI systems as strict-liability products legally and practically misguided, 2026.
Anjney Midha, a16z called liability for downstream misuse misguided, 2024.

Sources

Durbin, Hawley introduce bill allowing victims to sue AI companies (Sept 29, 2025)
S. 2937 bill status (govinfo): referred to Judiciary; no further action

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.