OpenAI backs liability limits
OpenAI supported a proposed Illinois bill that would narrow when AI firms can be held liable, even in cases of large-scale harm, signalling vendors are trying to limit legal exposure as enterprise sales expand. That lobbying shift adds urgency for customers to demand stronger contractual protections and clearer internal escalation paths when they buy AI services. (wired.com)
OpenAI did not just fight an AI bill this week. It showed up in Illinois to support one that would make it harder to sue AI model makers after extreme disasters, including cases involving 100 deaths or $1 billion in property damage. (wired.com) The bill is Illinois Senate Bill 3444, called the Artificial Intelligence Safety Act. It says a developer of a “frontier” model would not be liable for “critical harms” if the company did not act intentionally or recklessly and posted safety and transparency reports on its website. (ilga.gov) A “frontier” model here means an artificial intelligence system that cost more than $100 million in computing power to train. That threshold points at the biggest labs, including OpenAI, Google, Anthropic, Meta, and xAI, not the average software company. (legiscan.com) (wired.com) The key split is between the company that builds the model and the company that uses it. Illinois is debating whether the model maker should carry less legal risk while the hospital, bank, insurer, or defense contractor using the model carries more. (wired.com) That is a change in strategy for OpenAI. Wired reported that the company had mostly played defense before, trying to block state bills it disliked, while this time it backed a framework that gives large model vendors a clearer liability shield. (wired.com) OpenAI’s support came through testimony from its head of economic policy, Ronnie Chatterji. He argued that a patchwork of state rules could slow United States leadership in artificial intelligence and said Illinois should align with emerging federal approaches instead of creating broader state liability. (wired.com) The bill does not erase all responsibility. It still allows claims if a developer intentionally or recklessly caused the harm, but it raises the bar far above ordinary negligence, which is the standard many product cases use. (legiscan.com) (wired.com) Critics say that matters because posting a safety report is not the same thing as building a safe product. Scott Wisor of the Secure AI Project told Wired that Illinois voters strongly opposed exempting AI firms from liability and said existing companies should not get reduced exposure just because their systems are new. (wired.com) (capitolfax.com) Illinois is an important place for this fight because it already has a reputation for tougher tech rules, including the Biometric Information Privacy Act, a 2008 state law that created a private right to sue over misuse of faceprints and fingerprints. A liability shield passing there would give AI vendors a strong template to copy in other states. (wired.com) (ilga.gov) For companies buying artificial intelligence tools, this is the part to watch: if the model vendor gets statutory protection, the customer may be the easiest defendant left in court. That pushes the real negotiation into contracts, where buyers can demand indemnities, audit rights, incident reporting deadlines, and named escalation contacts before deployment. (wired.com)
