AI labs split on liability
OpenAI and Anthropic publicly disagree over a proposed Illinois AI liability bill — Anthropic says the measure, backed by OpenAI, would let labs evade responsibility for catastrophic harms, while OpenAI supports the text. The disagreement has exposed a deeper rift about how much downstream legal protection frontier-model developers should receive, with coverage highlighting the political and commercial stakes of the split. (wired.com) (techbuzz.ai)
OpenAI and Anthropic have split publicly over an Illinois bill that would limit when frontier artificial intelligence developers can be sued for catastrophic harm. (wired.com) The measure is Illinois Senate Bill 3444, filed on February 4, 2026 by State Senator Bill Cunningham. Its summary says developers of a frontier model would not be liable for “critical harms” if they did not act intentionally or recklessly and if they publish a safety protocol and transparency report. (ilga.gov) (legiscan.com) Wired reported on April 10 that OpenAI testified in favor of the bill, which defines “critical harms” to include events such as mass casualties or at least $1 billion in property damage. Five days later, Wired reported that Anthropic opposed the same proposal and argued it would shield labs too broadly. (wired.com 1) (wired.com 2) The fight is about a basic legal question: when a powerful model is misused downstream, how much responsibility stays with the company that built it. Senate Bill 3444 answers that question by giving developers a liability safe harbor if they meet the bill’s disclosure and conduct standards. (legiscan.com) (wired.com) The bill also lets companies satisfy the Illinois requirements by agreeing to European Union rules or by entering an agreement with a United States federal agency that meets specified conditions. It says the state law would stop applying if the federal government later adopts overlapping rules for frontier-model developers. (legiscan.com) That structure has turned a state bill into a proxy fight over national artificial intelligence policy while Congress has not passed a broad federal liability framework for frontier models. It also surfaces a commercial split between two companies that both sell large models but have taken different public lines on safety governance. (wired.com) (anthropic.com) (openai.com) Anthropic has spent the past year arguing for stricter rules on the most capable systems, including reporting and oversight for models that could enable severe misuse. OpenAI has also published safety frameworks, but its support for Senate Bill 3444 shows it is willing to back a liability regime that gives developers more protection if they follow stated safeguards. (anthropic.com) (openai.com) (wired.com) As of March 27, 2026, the Illinois Senate had set an April 24 committee deadline for the bill, according to LegiScan’s tracking page. The immediate question is not just whether Senate Bill 3444 advances, but whether other states borrow its safe-harbor model for artificial intelligence before Washington acts. (legiscan.com)
