OpenAI backs liability shield bill

OpenAI is backing an Illinois bill that would make it harder to sue the companies building the biggest artificial intelligence models when those models are tied to extreme disasters. Wired reported on April 10 that OpenAI testified for the bill, and the Illinois legislature lists it as Senate Bill 3444, the Artificial Intelligence Safety Act. (wired.com) (ilga.gov) The bill is aimed at “frontier” models, which is the policy term for the largest systems at the edge of current capability, like the difference between a family sedan and a Formula One car. Illinois says a developer of one of those models would not be liable for “critical harms” if it did not intentionally or recklessly cause them and if it publishes a safety and security protocol plus a transparency report on its website. (ilga.gov) Wired says the harms covered here are not ordinary chatbot mistakes like a bad homework answer or a fake legal citation. The bill is written around disasters on the scale of death or serious injury to 100 or more people, or at least $1 billion in property damage. (wired.com) That means the legal fight is moving to a narrower question: when does a model maker count as having caused the damage, and when does blame shift to the person or company that used the model. In plain terms, the bill treats the model builder more like a carmaker whose vehicle was misused than like the driver who hit the gas. (wired.com) (ilga.gov) Illinois is not writing this in a vacuum. The same legislature is also considering Senate Bill 3312, which would require large frontier developers to publish catastrophic risk, transparency, and cybersecurity frameworks and to report critical safety incidents to state agencies. (ilga.gov) Illinois is also considering Senate Bill 3590, an Artificial Intelligence Product Liability Act, which points in the opposite direction by laying out product-liability claims against developers for defective design, warnings, and warranties. In one statehouse, lawmakers are debating both a shield and a sword at the same time. (ilga.gov) OpenAI’s support matters because the company has spent much of the past two years arguing over rules written by other people, especially in California and Washington. This time, instead of just resisting a proposal, it is backing language that would define when a frontier lab can be sued after a catastrophe. (wired.com) (qz.com) The company has also spent the past year publicly describing misuse risks from its own systems, including state-linked influence operations and other malicious activity it says it disrupted. A bill like this turns those warnings into a legal question: if a lab says the danger is real, how much responsibility should it still carry when the danger shows up in the real world. (openai.com) (cdn.openai.com) The bigger surprise is where this is happening. Congress still has no comprehensive artificial intelligence liability law, so Illinois is trying to write one of the first real answers at the state level, and other states will be able to copy whichever version survives. (ilga.gov 1) (ilga.gov 2) If Senate Bill 3444 advances, the template is simple: publish safety paperwork, avoid intentional or reckless conduct, and get a liability shield when the damage is catastrophic. For the biggest artificial intelligence labs, that is not just a court argument anymore; it is a legislative strategy with a bill number attached. (ilga.gov)