llinois has put forth the latest attempt to regulate the rapid pace of model development.
On Monday, Illinois Governor JB Pritzker signed SB 315, a bill aiming to hold AI companies accountable for the powerful models they develop. The bill was passed unanimously in the Illinois state Senate and House in May, and was supported by both OpenAI and Anthropic.
"As AI systems become more powerful and the federal government is unwilling to step in, states have a responsibility to protect our people from the dangers of AI while still harnessing the unique potential of the technology," Pritzker said in a statement.
The bill, which is based on similar laws passed in New York and California in 2025, applies only to large frontier AI developers that generate $500 million or more in revenue. The bill will go into effect on January 1, 2027.
Though it targets a narrow audience, it applies strong safety regulation to some of the most powerful entities in AI, including:
- Requires these companies to publicly disclose their safety frameworks and take part in third-party evaluations of their safety practices
- Compels them to report their models' critical safety incidents, such as unauthorized theft, model compromise or user deception
- Assesses for catastrophic risks, such as loss of control, use to create weapons, or the ability to carry out large-scale cyberattacks
- Noncompliance can incur financial penalties, including a $1 million fine for the first violation and $3 million for any subsequent violations
Cesar Fernandez, head of US state and local government relations at Anthropic, said in a statement that the bill represents an "important step" toward the accountability that AI demands, as it pairs both transparency requirements with independent evaluations. "Anthropic is proud to have been the first AI lab to support this bill."
The bill comes as AI companies navigate a minefield of US government intervention in the release of their increasingly powerful models. After the rocky rollout of Anthropic's Mythos and Fable models, and OpenAI preemptively holding back its most recent model, GPT-5.6, the government's regulatory power over these companies' model releases is far from set in stone.
However, these companies may be working with the government on a solution: Last week, The Financial Times reported that many major AI firms are in talks with government officials to create a voluntary set of standards for releasing advanced models.
Our Deeper View #
What sets the Illinois bill apart from predecessors in California and New York is that it requires independent, third-party evaluation of model safety. Without an unbiased party to audit model behavior, users must place their trust in the internal evaluations and standards to which AI companies hold themselves. Additionally, SB 315 and similar bills go a step beyond any standards that the federal government could muster because they are required, rather than voluntary. While it's good that the federal government is trying to create a unified standard for model releases, these state laws go further by holding these companies legally accountable for any potential harm they cause.