Key facts
- Illinois Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act (Senate Bill 315).
- The law applies to AI models generating over $500 million annually and trained with massive computing power.
- Developers must publish AI frameworks assessing 'catastrophic risk' and report incidents of potential harm.
- The bill mandates annual third-party audits for AI model developers.
- Violations can result in civil penalties of up to $1 million for the first offense and $3 million for subsequent offenses.
Illinois Governor JB Pritzker has signed into law Senate Bill 315, the Artificial Intelligence Safety Measures Act, a landmark piece of legislation aimed at regulating artificial intelligence. This bill, modeled after similar laws in California and New York, seeks to establish state-driven guardrails for AI development and deployment in the absence of federal regulation.
Pritzker emphasized the need for legislative action, stating that many in Congress are influenced by special interests that benefit from a lack of regulation. He expressed a desire for thoughtful guardrails that serve both industry and the public, contrasting it with a scenario where a few actors evade accountability. The new law specifically targets the largest AI models, defined as those generating over $500 million in annual revenue and trained using substantial computing power.
The legislation introduces new reporting standards for AI models that could be used for large-scale harms, such as assisting in the creation of chemical or biological weapons or facilitating cyber-attacks. Senator Mary Edly-Allen highlighted the urgency for states to act, stating a reluctance to wait for federal action. The combined efforts of Illinois, California, and New York are expected to create a de facto national standard, representing approximately 40% of the U.S. AI market.
Developers will be required to publish an AI framework detailing how they identify and assess 'catastrophic risk,' defined as the potential for incidents causing death or serious injury to over 50 people or more than $1 million in property damage. They must also report such incidents within 72 hours, or 24 hours if an imminent risk is present. Representative Daniel Didech pointed to real-world examples, including an AI-inspired mass shooting and an attack on a water utility, to underscore the non-theoretical nature of these risks.
Illinois' version of the law includes a unique requirement for mandatory annual third-party audits, a provision that drew concern from industry groups like TechNet, who worried about subjective determinations without national standards. Despite these concerns, major AI developers like OpenAI and Anthropic supported the bill, which passed with broad bipartisan support. OpenAI, while preferring federal leadership, acknowledged the role states like Illinois, California, and New York can play in shaping AI frameworks.
Companies that violate the new law face civil penalties, starting at up to $1 million for a first offense and escalating to up to $3 million for subsequent violations, enforced by the attorney general's office. Lawmakers and advocates anticipate further work on AI regulation, with potential future evaluations focusing on risks in medical care and education.