OpenAI pushes liability shield
OpenAI backed an Illinois bill that would sharply limit when AI firms can be held liable, asking lawmakers to set a high bar for plaintiff claims even in cases of serious harm. At the same time, courts are tightening scrutiny of AI firms — a judge recently criticized a key OpenAI witness in the New York Times publishers’ copyright case for unreliable testimony, highlighting rising legal risk for labs. The clash between industry lobbying for liability protections and more skeptical judges suggests the next phase of AI policy will mix state bills and courtroom battles. (wired.com) (mercurynews.com)
OpenAI is now arguing in Springfield that an artificial intelligence company should usually not be on the hook even if its model is tied to a disaster on the scale of 100 deaths, 100 serious injuries, or $1 billion in property damage. The company backed Illinois Senate Bill 3444, a proposal called the Artificial Intelligence Safety Act. (wired.com) (ilga.gov) The shield in that bill is not absolute. It says a developer of a frontier artificial intelligence model would not be liable for “critical harms” if the developer did not act intentionally or recklessly and if it published a safety and security protocol plus a transparency report. (ilga.gov) (qz.com) Illinois picked an unusually high threshold for the word “critical.” The bill covers events like mass casualties, at least $1 billion in property damage, or use of artificial intelligence to help build a chemical, biological, radiological, or nuclear weapon. (qz.com) (wired.com) That is a different fight from the Illinois law that already took effect on January 1, 2026. House Bill 3773 amended the Illinois Human Rights Act so employers can be punished if artificial intelligence tools discriminate in hiring, promotion, discipline, or other job decisions. (natlawreview.com) (ogletree.com) So Illinois is becoming a two-track test case. One state law already treats artificial intelligence misuse at work like a civil-rights problem, while the new Senate bill would make it harder to sue the companies building the biggest models when the alleged harm comes later and at much larger scale. (natlawreview.com) (wired.com) At the same time, judges in New York are moving in the opposite direction by pressing OpenAI harder in copyright cases. The New York Times sued OpenAI and Microsoft on December 27, 2023, accusing them of copyright infringement tied to training and outputs. (courtlistener.com) This week, Magistrate Judge Ona T. Wang ordered a second deposition of an OpenAI witness after criticizing his testimony as too vague to rely on. Mercury News reported that the judge described the witness’s recollections as “hazy,” which is the kind of courtroom language that signals distrust, not deference. (mercurynews.com) (pacermonitor.com) OpenAI is also fighting a separate front in that same New York Times case over user data. On its own site, OpenAI says it was ordered to comply with a demand involving 20 million ChatGPT conversations, while saying it de-identified the data and kept access tightly controlled. (openai.com) Put those two tracks together and the picture gets clearer. In statehouses, OpenAI is asking lawmakers to write a liability rule that protects model makers unless plaintiffs can prove something close to deliberate or reckless conduct, while in federal court the company is facing judges who are demanding more precise evidence and more complete testimony. (wired.com) (mercurynews.com) That is where artificial intelligence policy is heading in 2026. The next rules are not coming from one big federal law, but from a patchwork of state bills like Illinois Senate Bill 3444 and court orders in cases like The New York Times v. Microsoft and OpenAI. (ilga.gov) (courtlistener.com)
