Illinois Senate passes bill requiring AI developers to disclose safety risks
The Illinois Senate voted 52-5 to advance legislation requiring major artificial intelligence developers to increase transparency and report on catastrophic risks. Senate Bill 315, announced May 13, targets companies like Meta, OpenAI, and Anthropic with revenues exceeding $500 million.
The bill requires these large developers to adopt a transparency framework, employ third-party auditors, and disclose a model's potential for catastrophic harm. It mirrors similar laws California and New York passed in late 2025, creating what lawmakers describe as a "de facto" national standard.
What companies must do
Under the bill, large "frontier" AI developers must create and publish a transparency framework explaining how they measure model capabilities, assess the chance of catastrophic risk, and identify and respond to safety incidents.
Three leading AI companies have reported in their own safety evaluations that their models could provide meaningful assistance in building biological weapons. Anthropic said one of its models was so powerful the company could not release it publicly.
OpenAI and Anthropic both testified in support, citing the need for a national standard focused on the "most capable" models and worst-case harms. This approach, they argued, allows smaller companies to innovate while protecting consumers.
The auditing requirement divides stakeholders
Illinois's bill goes further than California and New York by requiring third-party audits to ensure compliance. This provision became contentious during committee debate.
Jeremy Kudon, executive director of the American Innovators Network, called audits an "expensive requirement" for startups with few or no lawyers. Startups would need to use capital for legal and lobbying costs simply to comply, he said.
Senate Republican Leader John Curran questioned whether the bill defined auditors clearly enough. "I don't see anything that would prohibit my next-door neighbor from being a third-party auditor," he said in committee.
Amendments made this week clarified who can conduct audits, what audits should include, and how to protect proprietary information. The effective deadline was extended from 2027 to 2028.
Federal action spurred state regulation
Illinois lawmakers moved forward with regulation partly because of federal inaction. President Trump previously issued an order discouraging state regulation but has recently floated a second executive order directing developers to report powerful models to federal officials.
State Sen. Mary Edly-Allen sponsored the bill. The legislation is part of an eight-bill package addressing AI governance.
Concerns about keeping pace with technology
Opponents raised questions about whether the bill's definitions will hold as AI advances. The $500 million revenue threshold may become less relevant as companies grow faster and AI becomes more prevalent.
Model capabilities double every four months, according to Scott Wisor, policy director for Secure AI. Legislation takes time to pass, but AI development doesn't slow down.
Jonathan Iwry, a fellow at the Wharton Accountable AI Lab at the University of Pennsylvania, called the bill a positive development but noted it sets a low baseline. "The real question is whether these laws ultimately serve as a springboard toward stronger, more mature forms of AI governance, or instead allow what are relatively modest baseline accountability standards to get locked in as the long-term norm," he said.
Even companies developing frontier models say they cannot reliably predict or control their systems' behavior. Against that backdrop, Iwry said, requirements like transparency, testing, and auditing should be understood as foundational safeguards, not the upper limit of accountability.
Learn more about AI for Government and Generative AI and LLM regulation.
Your membership also unlocks:
