As tech CEOs and safety advocates press Washington for federal AI regulation, Congress appears unlikely to act. But as they did in recent years with data privacy, states are stepping up. A law approved this summer in Illinois is taking a demanding approach, which Dera Nevin and Kiley Kio of FTI Consulting unpack.
The current trajectory of AI regulation points toward a meaningful tightening in legal and operational accountability for large frontier developers, including for incident reporting and management of oversight mechanisms. To this end, Illinois Gov. JB Pritzker recently signed the Artificial Intelligence Safety Measures Act (SB 315) into law, set to take effect Jan. 1, 2027, with certain requirements activating the following year. While similar to California’s SB 53 and New York’s Responsible AI Safety and Education Act, which passed in late 2025, the act adds a mandatory annual third-party audit.
With a primary focus on large frontier model developers (encompassing “a person who trains, or initiates the training of, a frontier model using computing power that meets the technical specifications set forth in the definition,” with additional revenue thresholds), SB 315 and its implementation of independent third-party audits of AI frameworks and internal controls will push frontier model developers toward operationalizing governance programs and providing annual reporting to state regulatory oversight. It may also trigger more scrutiny from enterprise customers in associated procurement due diligence, reshaping vendor contracts with frontier model developers.
As SB 315 intensifies scrutiny over critical safety incidents and their management and disclosure, frontier developers will have direct incentives to understand, verify and potentially limit how their models are used downstream. These concerns will start to appear in enterprise agreements and corresponding procurement requirements. Enterprises that implement frontier models may face increasing scrutiny from frontier developers about their own governance documentation, risk classifications and safety incident reporting progress. For example, frontier model developers may be expected to require notification of catastrophic risk incidents from their customers within timelines that enable compliance with the regulatory reporting and notification requirements of SB 315. As these new incident reporting requirements come into force, this could impact model release timelines at the frontier model labs.
The law includes extra-jurisdictional, and potentially global, reach. Other included mechanisms, such as publication of frameworks and notices on the developer’s websites, will increase transparency on frontier models and identified catastrophic risks. This greater oversight is expected to have effects beyond Illinois, as public perceptions on model safety and effective AI governance and best practices evolve.
Key requirements
Among the requirements in SB 315, which are enforceable through defined monetary penalties and civil penalties, covered entities must file a disclosure statement in a proscribed form with the Illinois Emergency Management Agency and pay a required fee. Covered entities must also publish a transparency report that includes specified information about the frontier model and potential catastrophic risks on the company’s website, develop an AI framework, undergo annual third-party audits of that framework and publish the audit results, report critical safety incidents and maintain whistleblower protections for employees.
Large frontier developers must write, implement, comply with and clearly publish on their website an AI framework that describes how the frontier developer incorporates standards and industry-consensus best practices. Developers must also assess whether a frontier model has capabilities that could pose a catastrophic risk and show the processes that trigger revisions to the AI framework. The AI framework must be reviewed annually and updated as needed. Material modifications to AI frameworks must be disclosed on the frontier developer’s public-facing websites.
Another distinct feature of SB 315 is a mandatory third-party auditing requirement. Beginning Jan. 1, 2028, companies that qualify as large frontier model developers, a revenue-sized-based threshold, must annually retain an independent third-party auditor to evaluate compliance, including the assessments of catastrophic risks identified within the AI framework. The audit must include a description of whether the developer has complied with the requirements to prepare an AI framework, outline any material deviations from that framework and assess its internal controls. Additionally, audits must review the developer’s safety protocols and internal controls consistent with generally accepted auditing standards. The statute establishes this audit obligation ahead of state agency rulemaking, auditor licensing criteria or the identification of formalized testing methodologies, particularly for AI safety. Within 30 days of receiving that audit report, the frontier developer must publish a summary conspicuously on its website and submit a redacted version of the report to the Illinois attorney general and the Illinois Emergency Management Agency.
Additionally, large frontier developers must create documented technical and organizational protocols to manage, assess and mitigate capabilities that could pose catastrophic risks. The act sets a 72-hour reporting window to law enforcement and applicable agencies for critical safety incidents and specifies information that must be communicated. If a critical safety incident poses an imminent risk of death or serious physical injury, the reporting window shortens to 24 hours.
Such critical safety incidents include: unauthorized access, modification of model weights or loss of control of a model causing death or injury; use of a model for deceptive techniques to subvert developer control; or harm resulting from the materialization of a catastrophic risk. Activities like a frontier model providing expert-level assistance in the creation or release of certain weapons, evading control of the developer or undertaking cyberattacks or conduct that would constitute certain crimes if committed by a human are also defined as critical risks that must be reported. The focus on specified high-risk cases is unique compared to AI regulations established in other jurisdictions.
When critical safety incidents occur, reporting obligations fall on developers. Incident reports are currently confidential and may be exempt from Freedom of Information Act disclosure. In other words, Illinois is unlikely to serve as an immediate public resource for AI incident data, yet starting Jan. 1, 2029, the state will publish an annual aggregated and anonymized report of critical safety incidents that have been reviewed or reported.
Substantiate Your AI Claims Before They Become AI-Washing Challenges
With models, datasets and vendor APIs changing constantly, a claim that was accurate in the past may no longer hold up
Read moreDetailsSB 315 against a wider regulatory landscape
A wave of AI governance legislation has taken shape across the US and internationally, with each framework reflecting different legislative priorities. Among them, though, several consistent themes have emerged.
The EU AI Act remains the international benchmark for risk-based AI governance. California SB 53, New York’s RAISE Act and Illinois SB 315 converge on critical risks and developer obligations, with New York adding Department of Financial Services oversight. Other emerging AI governance legislation, such as Connecticut SB 5, takes a wider lens, covering employment systems, chatbots, synthetic content and youth safety. Some state legislation targets specific harm, such as the use of AI in employment screening or the delivery of healthcare services or the simulation of name, image or likeness of a living person. Together, these efforts signal an accelerating state legislative trend toward stricter AI regulation.
The requirements of SB 315 do not directly apply to organizations that acquire and implement frontier AI models but are still likely to impact how organizations use these models downstream. For example, the transparency and audit rules in SB 315 may be framed as industry benchmarks, so organizations implementing frontier models may be held to similar standards by their own customers and users. Organizations that implement frontier models may benefit from preparing governance infrastructure now, including evidence-based compliance programs, to demonstrate oversight and the capability of generating audit-ready documentation.
Organizations that implement frontier AI models should consider evaluating their frameworks to ensure they are equipped for autonomous systems, test them under realistic conditions, validate safety mechanisms and identify operational and security risks before implementation. Organizations should also create and review incident response protocols to ensure alignment with reporting timelines established by laws like SB 315, as these are likely to become embedded within their procurement contracts with frontier developers. Inventorying current AI implementations and assessing exposure under applicable frameworks, such as state-level legislation or standards bodies like NIST or ISO 42001, is a natural starting point.
Operationally, implementers of frontier models should pay close attention to frontier model application programming interfaces (APIs) and be aware that third-party auditing of frontier models’ safety frameworks will draw attention to, but also constrain, what their APIs expose. As frontier models incorporate elevated standards into procurement, implementers may also experience requests for evidence of testing protocols, incident detection protocols and logs, and red-teaming outputs, potentially both by frontier models and their own customers. Implementers may want to ensure procedures and documentation are in place by January 2028 to address these requests.
Dera Nevin
Kiley Kio
