State AI Safety Laws: California, New York, and Illinois

Provision

Text

CA
TFAIA

NY
RAISE Act

IL
AI Safety Act

Core Requirement

Frontier AI Framework: A set of documented technical and organizational protocols that a large developer must write, implement, and publish regarding how it will manage, assess, and mitigate catastrophic risks (as defined in the law; see “Risk” below).

●

●

●

Models

The law covers a:

• “Frontier model,” defined as a foundation model trained with over 10²⁶ operations.
• “Foundation model” defined as a model that is ALL of the following:
  • Trained on a broad data set
  • Designed for generality of output
  • Adaptable to a wide range of distinctive tasks.

●

●

●

Developers

The definition of a “developer” includes:

• “Frontier developer,” defined as “a person who has trained, or initiated the training of, a frontier model, with respect to which the person has used, or intends to use, at least as much computing power to train the frontier model” as required by the law.
• “Large frontier developer,” defined as a frontier developer that together with its affiliates had more than $500 million in annual gross revenue in the preceding year.

●

●

●

Developer explicitly excludes:

• Accredited colleges and universities engaged in academic research,
• The Empire AI consortium or institute as defined in New York law.

×

●

×

Risk

“Catastrophic risk,” is defined as a foreseeable and material risk that a frontier developer’s development, storage, use or deployment of a frontier model will materially contribute to death of/injury to >50 people, OR >$1 billion in damage arising from a single incident involving a frontier model that is:

1. Providing expert-level assistance for creating or releasing a chemical, biological, radiological or nuclear weapon
2. Engaging in cyberattacks with no meaningful human oversight, or conduct that would, if committed by a human, constitute murder, assault, extortion, or theft, including theft by false pretenses,
3. Evading the control of its frontier developer or user.

●

●

●

Lead State Agency or Agencies

California:

• Office of Emergency Services (CA OES) for reporting.
  • The CA Attorney General’s office may also transmit “reports of critical safety incidents and reports from covered employees.”
• Department of Technology for updating definitions, annually, based on developments in technology and national/international standards.
• CalCompute consortium for a public cloud initiative that will be created within the CA Government Operations Agency.

●

×

×

New York:

• “An office” to be designated within the NY Department of Financial Services tasked with implementation of the RAISE Act, including updating definitions based on developments in technology and national/international standards.

×

●

×

Illinois:

• Illinois Emergency Management Agency and Office of Homeland Security (IEMA-OHS) administers the Act in consultation with the Illinois Attorney General.

×

×

●

State Administration Fee

A large frontier developer must pay its pro rata share of the state's costs to administer the Act.

×

×

●

Incident Definition

“Critical Safety Incident,” defined to include events such as the unauthorized access to a model's weights that results in death or bodily injury, the materialization of a catastrophic risk, or a model using deceptive techniques to subvert its developer's controls.

●

●

●

Incident Reporting

California:

Report "critical safety incidents" to:

• The OES within 15 days
• Law enforcement or public safety agencies when incidents pose imminent risk of death or serious injury within 24 hours.

●

×

×

New York:

Report "critical safety incidents" to:

• The DFS office within 72 hours
• Law enforcement or public safety agencies when incidents pose imminent risk of death or serious injury within 24 hours injury.

×

●

×

Illinois:

IEMA-OHS, in consultation with the Attorney General, must establish a mechanism that frontier developers and members of the public may use to report critical safety incidents to:

• IEMA-OHS and the Illinois Attorney General within 72 hours
• Law enforcement or public safety agencies when incidents pose imminent risk of death or serious physical injury within 24 hours.

×

×

●

Public Disclosure

Must publish the full "Frontier AI Framework" and a "Transparency Report" for each new model.

●

●

●

A large frontier developer may not develop, deploy, or operate a frontier model without filing a current disclosure statement with IEMA-OHS, and IEMA-OHS must publish a list of large frontier developers that have filed disclosure statements.

×

×

●

Authorizes a $1,000-per-day civil penalty, plus recovery of unpaid assessments, for failure to file the required disclosure statement or correct false disclosure information.

×

×

●

Third-Party Review

• The developer's Frontier AI Framework must describe its approach to using third parties for risk assessment.
• The Transparency Report for a model must state the extent to which third-party evaluators were involved.

●

●

●

Primary Penalties

The state’s Attorney General’s office may bring a civil action to enforce the law against violations.

●

●

●

• Up to $1 million per violation.

●

×

×

• Up to $1 million for the first violation.
• Up to $3 million for subsequent violations.

×

●

●

Whistleblower Protections

• Protects "covered employees," defined as those responsible for assessing, managing, or addressing risk of critical safety incidents, for reporting to an authority.
• Requires an anonymous internal reporting process.

●

×

●

An employee may sue for injunctive relief and be awarded reasonable attorney's fees.

●

×

×

Requires a process that provides monthly status updates to the reporting covered employee and shares disclosures and responses with officers and directors at least quarterly, subject to an exception for implicated officers or directors.

×

×

●

Scope

Explicitly applies to any frontier model developed, deployed, or operating, in whole or in part, within the state.

●

●

●

Severability Clause

●

●

●

Deference/ Preemption

Permits frontier developers to comply with federal laws/guidance regarding critical incident reporting provided the federal laws/guidance are intended to assess, detect, or mitigate catastrophic risk and the requirements for reporting are the same or more stringent than the state law.

●

●

●

Preempts local laws or ordinances passed after January 1, 2025, that regulate the same subject.

●

×

×

Makes regulation of frontier AI models an exclusive state power, prohibiting local governments from regulating frontier AI models as a “denial and limitation of home rule.”

×

×

●

Prohibition
of False Statements

Explicitly prohibits developers from making materially false or misleading statements about risk or compliance.

●

●

●

Confidentiality

Reporting of critical safety incidents and risk assessments submitted to OES are exempt from the California Public Records Act to protect trade secrets and public safety.

●

×

×

Exempts the following from disclosure under the state public records law:

• Critical safety incident reports.
• Internal-use catastrophic risk assessments.
• Unredacted third-party audit reports.
• Materials, work papers, notes, and derivative documents prepared by a third party in connection with an audit from disclosure under the state public records law, when those records are in the possession of IEMA-OHS or the Illinois Attorney General.
• Covered employee reports.

×

×

●

Annual State Report

The relevant state office will produce an annual public report containing anonymized and aggregated information from critical safety incident reports it has reviewed...

…beginning on January 1 of the year after the law goes into effect

●

●

×

…beginning on January 1, 2029

×

×

●

Independent Audit

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

×

×

●

Within 30 days after receiving the audit report, the large frontier developer must publish a high-level summary of the audit findings and a redacted copy of the report and transmit the redacted copy of the report to IEMA-OHS and the Illinois Attorney General.

×

×

●

The large frontier developer must retain an unredacted copy of the audit report for as long as the frontier model is deployed plus 5 years.

×

×

●

Internal Use Reporting

Requires large developers to transmit summaries of catastrophic risk assessments from the internal use of their models to the relevant state office every three months.

●

●

●