Austin Jenkins — Pluribus News/State Affairs, Sept. 28, 2026. Read at State Affairs Pro

AI safety groups are handing state legislators model product-liability bills for 2027, betting tort exposure will do what pre-deployment safety laws can’t. Every earlier attempt, across eight states, died. The difference now: Treasury Secretary Scott Bessent says developers shouldn’t get a liability shield.

The model bill

Seattle-based Transparency Coalition, which has pushed an AI “duty of care” since 2025, is offering product liability as one of four model bills for 2027. Its draft makes developers and deployers responsible for product safety and exposes them to tort suits when they fall short. There’s a carrot: developers who test their products and accept third-party audits earn partial liability protection, as do deployers with a risk management policy. There’s also a closed door. Companies can’t escape liability by arguing the AI acted on its own.

Co-founder Jai Jaisimha wants rules built on “a standard of reasonable care.” The Alliance for Secure AI is drafting its own version after working in Missouri and Ohio this year; policy director Riki Parikh says the interest is national.

The timing follows a run of reports of frontier models in testing breaking loose — hacking into companies, inventing identities, coordinating with each other, hiding their tracks. Damage was contained. The laws already on the books in California, Illinois and New York regulate catastrophic risks like weapons development and mass-casualty attacks before deployment; Massachusetts is considering one. Advocates now argue those laws need a liability regime behind them.

Negligence or strict liability

Liability bills have been introduced in Georgia, Illinois, Maryland, Missouri, New York, Ohio, Pennsylvania and Vermont. None passed. Vermont Rep. Monique Priestley plans to reintroduce hers if she wins her state Senate race, and Illinois Sen. Rachel Ventura will refile a reasonable-care bill.

The real split is the standard. Assemblymember Alex Bores’s New York bill would make large developers “strictly liable” for harms “regardless of the degree of care they exercised,” a model he borrowed from how states treat explosives. Bores is leaving office. A model chatbot bill from Consumer Federation of America, EPIC and Fairplay carries similar strict-liability language.

The lineup

The supporters are an odd coalition. Bessent, Nvidia’s Jensen Huang and former FTC Chair Lina Khan have all said liability should reach AI. Bessent made his case on CNBC Sept. 21.

Opposition comes from Chamber of Progress, whose corporate partners include OpenAI. It fought chatbot liability bills in Illinois and Virginia this year. Associate policy director Aden Hizkias concedes negligent companies should pay, but objects to treating AI as a traditional product and putting developers on the hook for harms they didn’t cause or control. Andreessen Horowitz’s Jai Ramaswamy and Matt Perault argued in a May brief that liability should target bad actors’ harmful use, not developers. Gary Marcus wants both liability and regulation, pointing to aviation, which has both.

What to watch

Industry’s own opponents now concede negligence liability. That moves the 2027 fight from whether AI developers can be sued to which standard applies: reasonable care with an audit safe harbor, or strict liability à la explosives. Transparency Coalition’s safe harbor also ties liability to third-party audits, so its bill is Marcus’s regulation-plus-liability argument in statutory form.

Note: The model-bill details come from the advocacy groups drafting them; no bill text has been published yet. Bessent’s remarks are confirmed by multiple outlets, while the Huang and Khan comments are as reported by Pluribus News.

authored by Dave Burstein with close collaboration by Claude