AI liability showdown
The AI industry is pushing for legal safe harbors: OpenAI has publicly backed an Illinois bill that would limit when AI firms can be sued for so‑called “critical harms.” ( ). That move comes as lawsuits and safety concerns multiply — a Prism News complaint alleges OpenAI ignored multiple warnings about a ChatGPT user flagged for mass‑casualty weapons activity, and creators’ copyright suits targeting Amazon, OpenAI and others have pushed the total U.S. AI copyright cases past 40. ( ). Those legal pressures are feeding a federal‑vs‑state fight over AI rules, with opinion pieces urging that states should not be pre‑empted while Congress stalls. ( )
OpenAI just backed an Illinois bill that would make it much harder to sue an artificial intelligence company even after an extreme disaster, including the death or serious injury of 100 or more people or at least $1 billion in property damage. The bill is Illinois Senate Bill 3444, called the Artificial Intelligence Safety Act. (wired.com, ilga.gov) The protection is not blanket immunity. Senate Bill 3444 says a developer of a “frontier” model would avoid liability only if it did not intentionally or recklessly cause the harm and if it published a safety and security protocol plus a transparency report on its website. (ilga.gov, qz.com) The bill is aimed at the biggest systems, not every chatbot on the internet. Reporting on the measure says it covers “frontier” models, which the proposal ties to very large training runs, including systems trained with more than $100 million in compute. (wired.com, letsdatascience.com) The list of harms in the bill is built around nightmare scenarios. It includes chemical, biological, radiological, or nuclear weapons, and it also covers situations where a model acts on its own in a way that would be a crime if a human did it and that leads to mass casualties or billion-dollar damage. (qz.com, theoutpost.ai) OpenAI’s argument is that companies need one clear rulebook instead of 50 different ones. Wired reported that the company testified for the Illinois bill while also pushing for a national framework that would avoid a state-by-state patchwork. (wired.com, qz.com) That push is landing at the exact moment the lawsuits are getting more concrete. On April 10, 2026, TechCrunch reported on a new suit by a woman identified as Jane Doe who says OpenAI ignored three warnings that her ex-boyfriend was dangerous, including an internal flag tied to “mass-casualty weapons” activity. (techcrunch.com, prismnews.com) Prism News had already reported a related timeline in the Tumbler Ridge case. It said OpenAI banned Jesse Van Rootselaar’s ChatGPT account in June 2025, months before the February 2026 killings in British Columbia, and contacted the Royal Canadian Mounted Police only after the shooting. (prismnews.com) The pressure is not only about physical harm. Copyright fights are piling up too, and CNET reported this week that YouTube creators sued Amazon over claims that its video model was trained on scraped videos, calling it one of dozens of cases testing how artificial intelligence companies gathered training data. (cnet.com, bakerlaw.com) Law firms and case trackers now describe the United States docket as a growing map of unresolved questions, not a few isolated complaints. BakerHostetler and Norton Rose Fulbright both say generative artificial intelligence litigation now spans copyright, class actions, training data, and ownership disputes across multiple companies. (bakerlaw.com, nortonrosefulbright.com) That is why this Illinois fight is bigger than one state bill. If courts are about to decide who pays when an artificial intelligence system causes harm, the industry wants the answer written now, while critics want states to keep writing stricter rules because Congress still has not passed a comprehensive federal artificial intelligence law. (cloudsecurityalliance.org, ropesgray.com) The federal government is moving in the opposite direction from many state advocates. Legal analyses of the White House framework released on March 20, 2026 say it pushes Congress toward a unified national approach and puts preemption of state laws at the center of the next artificial intelligence policy battle. (ropesgray.com, forbes.com) So the real question under this Illinois bill is simple. When a model is powerful enough to help cause a catastrophe, should publishing a safety report protect the company that built it, or should that question stay open for judges, juries, and state lawmakers. (ilga.gov, wired.com)
