OpenAI backs liability shield

OpenAI did not just fight a bill this week. It backed an Illinois Senate bill, Senate Bill 3444, that would make it much harder to sue the biggest artificial intelligence labs when their systems are tied to disasters on a very large scale. (ilga.gov) (qz.com) The bill says a lab would not be liable for “critical harms” if it did not act intentionally or recklessly and if it posted a safety and security protocol plus a transparency report on its website. Quartz reported on April 10, 2026 that OpenAI supported that approach. (ilga.gov) (qz.com) Illinois defines “critical harms” in Senate Bill 3444 with very high thresholds: death or serious injury of 100 or more people, at least $1 billion in property damage, or use of artificial intelligence to help create a chemical, biological, radiological, or nuclear weapon. Those are not edge-case typos in a draft; they are the bill’s core examples. (ilga.gov) (qz.com) The protection is aimed at “frontier” models, which the bill describes as models trained with more than $100 million in compute costs. That cutoff points at a small club of companies, including OpenAI, Google, Anthropic, Meta, and xAI, rather than the average software startup. (ilga.gov) (qz.com) That is why this fight is not really about chatbots making silly mistakes. It is about who pays when a very expensive model is linked to something closer to an industrial accident, a weapons incident, or a market-scale catastrophe. (ilga.gov) (wired.com) OpenAI’s public argument is that rules should focus on the most advanced systems while avoiding a 50-state patchwork. Wired reported that OpenAI spokesperson Jamie Radice said the company supports approaches that reduce serious harm risk and move toward “clearer, more consistent national standards.” (wired.com) The timing matters because courts and legislatures in the United States still have not settled a basic question: when an artificial intelligence model causes harm, is the model maker more like a product manufacturer, a publisher, or a toolmaker? Illinois is now one of the places where that answer is being written in real bill text instead of law review articles. (ilga.gov) (wired.com) The bill also gives labs more than one way to qualify for protection. Legislative summaries say a developer is deemed compliant if it follows certain European Union safety requirements or enters into a qualifying agreement with a federal agency, which means state liability limits could effectively be unlocked through other regulatory systems. (legiscan.com 1) (legiscan.com 2) Critics quoted by Wired say that flips the usual logic of liability law. In most industries, the bigger the possible harm, the stronger the incentive to prove you took every reasonable precaution; here, critics argue, posting reports and avoiding recklessness could become enough to block many lawsuits after the worst outcomes. (wired.com) This also looks like a shift in OpenAI’s strategy. Wired reported that the company had mostly been trying to stop stricter state proposals before, while this Illinois bill shows it supporting a model law that could become a template for other states or for Congress. (wired.com) (capitolfax.com) If Senate Bill 3444 or bills like it spread, the next big artificial intelligence policy fight will not be over whether labs should publish safety plans. It will be over whether those plans are a seat belt before a crash or a legal shield after one. (ilga.gov) (wired.com)