Illinois Turns Frontier AI Safety Programs Into Auditable Law

Illinois has moved frontier-model safety from voluntary policy statements into an auditable legal framework.

Governor JB Pritzker approved SB 315 as Public Act 104-0538, the Artificial Intelligence Safety Measures Act, on July 6, 2026. The Act takes effect January 1, 2027. It creates disclosure filings, model transparency reports, critical-safety-incident reporting, whistleblower protections, and—beginning in 2028—published safety frameworks and annual independent audits for the largest frontier developers.

Most small businesses will not be directly regulated as frontier-model developers. That does not make the law irrelevant to them. The records Illinois requires from large labs—model reports, catastrophic-risk assessments, incident processes, audit evidence, and governance ownership—are the same records enterprise buyers, insurers, investors, and regulators are likely to request from the AI supply chain.

The practical question is therefore two-part: Are you directly in scope, and what evidence should you expect from an upstream model provider even if you are not?

The Law Targets Frontier Models, Not Ordinary AI Features

The Act defines a frontier model as a broadly capable foundation model trained using more than 10^26 integer or floating-point operations. The threshold includes the original training run plus subsequent fine-tuning, reinforcement learning, or other material modifications.

A “large frontier developer” is a frontier developer that, together with its affiliates, had more than $500 million in annual gross revenue during the preceding calendar year.

Those definitions matter. A small company using an API, building a chatbot on top of an existing model, or fine-tuning a modest open model is not automatically a frontier developer. Nor does every AI vendor cross the large-developer revenue threshold. Teams should document the role they actually play rather than labeling every generative-AI product “frontier AI.”

January 2027: Filing, Transparency, and Incident Reporting Begin

The Act takes effect January 1, 2027. From that date, a large frontier developer may not develop, deploy, or operate a frontier model, in whole or in part in Illinois, without a current disclosure statement filed with the Illinois Emergency Management Agency and Office of Homeland Security and payment of the required regulatory fee. The filing must be renewed annually and when ownership or material reported information changes.

The Act also requires a frontier developer to publish a transparency report before or when deploying a new frontier model or a substantially modified version. The report must identify basic model facts such as release date, supported languages and output modalities, intended uses, and generally applicable use restrictions. Large developers must add summaries of catastrophic-risk assessments, results, third-party evaluator involvement, and other safety steps.

Critical-safety-incident reporting is another operational requirement. A frontier developer must report a qualifying incident to the Agency and Illinois Attorney General within 72 hours after learning facts sufficient to support a reasonable belief that the incident occurred. If the developer discovers an imminent risk of death or serious physical injury, it must disclose the incident within 24 hours to an appropriate authority.

These are escalation deadlines, not ordinary customer-support targets. Developers need a written decision path for identifying a potentially reportable event, involving legal and technical owners, preserving evidence, and meeting the clock without waiting for a complete root-cause analysis.

January 2028: Published Frameworks and Independent Audits

Beginning January 1, 2028, large frontier developers must write, implement, follow, and publicly post a frontier AI framework. The framework must address how the developer:

• ●Uses national standards, international standards, and industry-consensus practices.
• ●Defines and assesses capability thresholds associated with catastrophic risk.
• ●Applies mitigations and reviews them before model deployment or extensive internal use.
• ●Uses third parties to assess catastrophic risk and mitigation effectiveness.
• ●Secures unreleased model weights against unauthorized modification or transfer.
• ●Identifies and responds to critical safety incidents.
• ●Assigns internal governance responsibility and manages risks from internal model use.

The Audit Requirement Has Teeth

Beginning January 1, 2028—or 90 days after first qualifying as a large frontier developer, whichever is later—a covered developer must annually retain an independent third party to audit compliance with the framework requirements.

The auditor must have demonstrated competence, follow generally accepted auditing standards and best practices, and remain free of specified financial conflicts. The developer must give the auditor access to reasonably necessary materials. The resulting report must address compliance, material deviations, internal controls, methodology, personnel, conflicts, and the lead auditor’s certification.

The developer must retain an unredacted report while the relevant model remains deployed plus five years. Within 30 days after receiving the report, it must publish a high-level summary and a suitably redacted copy and transmit the redacted report to the Agency and Attorney General.

That is more than an annual policy review. It requires audit-ready evidence: versioned frameworks, assessment records, deployment approvals, incident logs, control ownership, evaluator records, and proof that published claims match internal practice.

Whistleblowers and Public Claims Are Part of the Control System

The Act prohibits frontier developers from blocking or retaliating against covered employees who disclose qualifying public-safety risks or violations. It also requires notice of those rights.

Separately, a frontier developer may not make a materially false or misleading statement about catastrophic risk or its management. A large developer may not make a materially false or misleading statement about implementing or complying with its frontier AI framework. Good-faith statements that were reasonable under the circumstances receive protection.

The governance lesson is straightforward: a polished safety page creates risk if the internal evidence does not support it. Claims review, employee escalation, and audit preparation belong in the same program—not in separate legal, communications, and engineering silos.

Penalties Focus on Missing Evidence and Broken Processes

A large frontier developer can face a civil penalty of up to $1 million for a first violation and up to $3 million for each subsequent violation for specified failures, including missing required documents, prohibited statements, failure to obtain the independent audit, failure to report a critical safety incident, or failure to follow its own frontier AI framework.

Operating without the required disclosure filing, submitting false filing information, or failing to pay an assessment can also produce a $1,000-per-day filing penalty plus unpaid assessments. Enforcement of the main civil penalties belongs exclusively to the Illinois Attorney General, and the Act does not create a private right of action.

Illinois also allows an interoperability route: the state may designate federal laws, regulations, or guidance that impose substantially equivalent or stricter requirements. A developer that properly declares reliance on a designated federal regime may be treated as compliant to that extent—but failure to meet the designated federal standard then becomes an Illinois violation. This is coordination, not a blanket exemption.

What Smaller AI Buyers and Builders Should Do

Most AIRegReady readers are unlikely to train a model above the compute threshold or exceed the large-developer revenue threshold. Their immediate work is vendor governance, not building a frontier-lab compliance department.

Add these questions to material model-provider reviews:

• ●Scope: Does the provider classify any model you use as a frontier model under Illinois, New York, California, or another jurisdiction?
• ●Model evidence: Can it provide a current model or system card, deployment transparency report, evaluation summary, and intended-use restrictions?
• ●Incident terms: What safety incidents trigger customer notice, regulatory reporting, suspension, or a forced model migration?
• ●Framework and audit: Does the provider publish a safety framework, and will independent audit summaries or reports be available?
• ●Change management: How will customers learn about substantial model modifications, altered restrictions, or new risks?
• ●Fallback: Can your organization switch models or disable a high-risk feature if an incident, enforcement action, or audit finding interrupts access?

The Bottom Line

Illinois is not regulating every chatbot or small AI business as a frontier lab. It is creating a documentation-and-assurance system for developers operating at the highest compute and revenue levels.

The direct deadlines belong to frontier developers: filings and incident processes in 2027, then published safety frameworks and annual audits in 2028. The indirect effects will travel faster. Customers will ask for model reports, assessment evidence, incident commitments, and audit results because the law turns those materials into recognizable proof of responsible operation.

If your company builds frontier models, start assigning owners now. If it buys them, start asking for the evidence now. The statute may regulate the lab, but the model risk still reaches the customer.

Related Regulations

Sources & References