Frontier AI Safety Laws: SB 53, RAISE Act, SB 315 (2027)

As of November 4, 2026, by our count of enacted laws, and subject to the search limits shown in the table on our AI laws hub, three states have frontier-model safety statutes: California's SB 53 (Bus. & Prof. Code 22757.10 ff.) is in force, and New York's RAISE Act (GBL article 44-B, as replaced by S8828) and Illinois SB 315 (P.A. 104-0538) are scheduled to take effect January 1, 2027.

Status as of November 4, 2026: SB 53 has been in force since January 1, 2026. The RAISE Act and SB 315 are signed but not yet in force. Illinois's framework duty starts January 1, 2028, and its audit duty starts on the later of that date or 90 days after a developer first qualifies. We found no federal statute in the public laws of the 119th Congress that generally preempts or pauses state AI laws, and the federal proposals that address frontier AI laws are not law.

This page compares state statutes aimed at developers of the largest AI models. It is general legal information, not legal advice. The California, New York and Illinois pages carry each state's fuller detail.

SB 53 vs. RAISE Act vs. SB 315: Comparison Table

The cells rest on the enacted texts linked in the header row, with section numbers so you can check them.

In all three states, the Attorney General's civil penalty reaches large frontier developers only, including for a missed incident report, even though every frontier developer must report. That gap is in the text of each law, and it is easy to miss in summaries that describe the reporting duty and the penalty together. Each law also says its duties are cumulative with other law and do not limit rights or remedies under existing law (SB 53 sec. 5(c); 1424; Sec. 30).

What "Frontier Model" and "Frontier Developer" Mean in Each Law

The three texts use nearly identical definitions, and the sponsor's memo on New York's S8828 says the changes align the act "more closely with California's SB 53." Coverage depends on stacked terms, so it helps to read them in order.

Foundation model. All three define this as an artificial intelligence model that is trained on a broad data set, designed for generality of output, and adaptable to a wide range of distinctive tasks. All three elements are required.

Frontier model. A foundation model "that was trained using a quantity of computing power greater than 10^26 integer or floating-point operations." The count "shall include computing for the original training run and for any subsequent fine-tuning, reinforcement learning, or other material modifications the developer applies to a preceding foundation model."

Frontier developer. Here the wording differs slightly. California and New York cover a person who "has trained, or initiated the training of," a frontier model, where the person "has used, or intends to use," at least that much computing power. Illinois covers a person "who trains, or initiates the training of, a frontier model using computing power that meets the technical specifications," and it has no "intends to use" clause. In each text, the definition turns on training.

Large frontier developer. A frontier developer whose gross revenues, counted together with its affiliates, passed $500 million in the preceding calendar year. Revenue, not model size, decides who owes the heavier duties.

Catastrophic risk. All three use a foreseeable and material risk that a frontier model will materially contribute to the death of, or serious injury to, more than 50 people, or more than $1 billion in damage to or loss of property, from a single incident. The incident must involve the model giving expert-level help with a chemical, biological, radiological or nuclear weapon, acting without meaningful human oversight in a cyberattack or conduct that would be murder, assault, extortion or theft if done by a human, or evading the control of its developer or user.

The definitions exclude information otherwise publicly accessible in a substantially similar form from a source other than a foundation model, lawful activity of the federal government, and harm from a model combined with other software if the model did not materially contribute. Loss in the value of equity does not count as damage to property (22757.16; 1423; Sec. 25(c)).

Critical safety incident. All three list four events:

• Unauthorized access to, modification of, or exfiltration of model weights that results in death or bodily injury.
• Harm resulting from the materialization of a catastrophic risk.
• Loss of control of a frontier model causing death or bodily injury.
• A frontier model using deceptive techniques to subvert its developer's controls or monitoring, outside an evaluation designed to elicit that behavior, in a manner that demonstrates materially increased catastrophic risk.

Each law also tells its agency to set up a way for a frontier developer or a member of the public to report an incident (22757.13(a); 1422(1); Sec. 15(a)).

California SB 53: The Transparency in Frontier AI Act (TFAIA)

Governor Newsom signed SB 53 on September 29, 2025, and it took effect January 1, 2026. It followed SB 1047, which he vetoed on September 29, 2024.

What every frontier developer owes. Before or when deploying a new or substantially modified frontier model, it publishes a transparency report listing its website, a contact mechanism, the release date, languages, output modalities, intended uses and generally applicable restrictions. It reports critical safety incidents to the Office of Emergency Services within 15 days of discovery. It may not make a materially false or misleading statement about catastrophic risk, with an exception for statements made in good faith that were reasonable under the circumstances.

What large developers add. A large frontier developer must write, implement, comply with and publish a frontier AI framework, review it at least once a year, and publish a material change within 30 days. Its transparency reports add summaries of catastrophic-risk assessments, their results and the extent of third-party evaluators' involvement. It also sends internal-use risk summaries to the Office of Emergency Services.

Enforcement. The penalty depends on severity and "does not exceed one million dollars ($1,000,000) per violation." It covers a large developer that fails to publish or transmit a compliant document, makes a barred statement, fails to report an incident, or fails to comply with its own framework. It is "recovered in a civil action brought only by the Attorney General."

Whistleblowers. Labor Code 1107.1 protects "covered employees," defined as those "responsible for assessing, managing, or addressing risk of critical safety incidents." A frontier developer may not stop or retaliate against their disclosures to the Attorney General, a federal authority, a person with authority over the employee, or another covered employee with authority to investigate, where the employee has reasonable cause to believe the developer's activities pose a specific and substantial danger to public health or safety from a catastrophic risk, or that it violated the chapter.

The Labor Code chapter has its own definitions (Lab. Code 1107). Its catastrophic-risk definition refers to foundation models, and its critical safety incident list adds damage to or loss of property to the model-weights event.

Large developers must provide an anonymous internal process with monthly status updates to the person who disclosed. A covered employee can bring a civil action under the section, and a court may award reasonable attorney's fees to a successful plaintiff.

What SB 53 does not do. It requires no annual third-party audit; the text refers only to using third parties in assessing risk. Two 2026 California laws build machinery for outside checks without requiring it: SB 813 on independent verification organizations and AB 1405 on an AI auditor registry. Executive Order N-9-26 asks for recommendations on possible amendments, including onsite audits and a "kill switch." Those are ideas under study, not law. Our California page covers them.

SB 53 preempts city and county rules adopted on or after January 1, 2025, specifically related to frontier developers' management of catastrophic risk. By its own terms, the act "shall not apply to the extent that it is preempted by federal law."

New York RAISE Act (S8828, GBL Article 44-B): What Replaced the 2025 Text

The RAISE Act has two versions, and only the second is scheduled to take effect. Governor Hochul signed the first, S6953-B, as chapter 699 of 2025 on December 19, 2025. She signed S8828, chapter 96 of 2026, on March 27, 2026. Its sponsor memo says it exists "to repeal the underlying chapter and establish a new regulatory regime."

Section 2 of S8828 repeals article 44-B as first enacted and adds a new article 44-B. Section 3 replaces the 90-day effective date with January 1, 2027. Some older commentary describes the first version, so this table shows what changed.

Scope and exceptions. Article 44-B "shall only apply to frontier models that are developed, deployed, or operating in whole or in part in New York state" (1425). It does not apply to accredited New York colleges and universities to the extent they are doing academic research on AI models, or to the Empire AI consortium or institute (1426).

The office. The statute defines it as "an office within the department of financial services, which shall report to the superintendent of financial services and is tasked with implementation of this article" (1420(16)). It does not name the office.

Section 1429 authorizes the office to adopt rules. We have not confirmed whether it has adopted a rule or opened a filing process.

Disclosure statement. Section 1428 bars a large frontier developer from developing, deploying or operating a frontier model, in whole or in part in New York, without a current disclosure statement on file and its share of costs paid.

The 72-hour clock. It runs from a determination that an incident occurred, or from learning facts sufficient to establish a reasonable belief that one did. These reports are exempt from disclosure under the public officers law (1422(6)). Starting January 1, 2028, the office produces an annual report with anonymized and aggregated incident information and sends it to the Governor and legislative leaders (1422(7)).

Enforcement. The Attorney General's penalty applies where a large frontier developer fails to publish or transmit a compliant document, makes a barred statement, fails to report an incident, or fails to comply with its own framework (1427(1)). Separately, the office may, after notice and hearing, impose $1,000 for each day a disclosure is not filed or false information is not corrected, plus an amount equal to the assessments owed (1428(5)). Article 44-B creates no private right of action (1427(2)). Our New York page has the full duty list.

Illinois SB 315: The Artificial Intelligence Safety Measures Act

The Illinois General Assembly passed SB 315 in May 2026, and the Governor approved it July 6, 2026, as P.A. 104-0538. It takes effect January 1, 2027. We have not seen an ILCS cite for it yet. Its duties do not all start the same day.

January 1, 2027. The 72-hour incident duty covers every frontier developer from this date; it is not phased to 2028 and it is not limited to large developers. Large frontier developers also need a disclosure statement on file with the Agency and the required fee paid before developing, deploying or operating a frontier model in whole or in part in Illinois (Sec. 18). "Agency" means the Illinois Emergency Management Agency and Office of Homeland Security.

No separate date stated. Transparency reports (Sec. 10(c)), the false-statement bar (Sec. 10(f)), internal-use summaries (Sec. 10(e)) and whistleblower protections (Sec. 20) state no start date of their own, and the act takes effect January 1, 2027.

January 1, 2028. Large frontier developers must write, implement, comply with and publish a frontier AI framework (Sec. 10(a)).

The third-party audit. Sec. 10(d) reads: "Beginning on January 1, 2028 or 90 days after a developer first qualifies as a large frontier developer, whichever is later, a large frontier developer shall annually retain a third party to perform an independent audit of compliance with the requirements of this Section." The audit tests Section 10, not the whole act.

The developer keeps an unredacted report "for as long as a frontier model is deployed plus 5 years." Within 30 days of receiving it, the developer publishes a high-level summary and a redacted copy, and sends the redacted copy to the Agency and the Attorney General.

Enforcement. Section 25(a) reaches a large frontier developer that fails to publish or transmit a compliant document, makes a barred statement, fails to have the audit done, fails to report an incident, or fails to comply with its own framework. The action is "brought exclusively by the Attorney General," and "Nothing in this Act shall be construed to establish a private right of action associated with violations of this Act." Separately, the Agency may, after notice and hearing, impose the Sec. 18(e) filing penalty of $1,000 a day, plus an amount equal to the assessments owed.

Local rules. Section 35 says the regulation of AI frontier models "is an exclusive power and function of the State," a limit on home rule.

Our Illinois page has the full table of duties and the Whistleblower Act detail.

Connecticut: A Frontier Whistleblower Provision Only

Connecticut has a frontier whistleblower provision only. PA 26-15 sec. 2, in force since October 1, 2026, protects employees of frontier developers, with added protections for covered employees. By January 1, 2027, each large frontier developer must set up an internal process for covered employees to report anonymously to the company. It is not an incident report to the state. We do not count it among the three safety statutes.

Frontier AI Bills That Failed or Are Pending

These are the frontier bills our state research surfaced. This list is not exhaustive; the hub table shows the search limits by state.

Do Federal Preemption Efforts Affect SB 53, RAISE and SB 315?

As of November 4, 2026, we found no federal statute in the public laws of the 119th Congress that generally preempts or pauses state AI laws, and none that sets frontier-model safety duties on private developers generally. Federal policy on state AI laws runs through executive orders, agency action, litigation and proposals to Congress.

Executive Order 14365 directs agencies to challenge or discourage state AI laws the administration views as conflicting with national policy. It does not itself declare any state law preempted. The White House's March 2026 recommendations to Congress say "States should not be permitted to regulate AI development." That is a recommendation, not law.

Our preemption tracker lists the federal bills we follow, including a broader moratorium bill. Two proposals are aimed most directly at frontier laws:

• Great American AI Act (discussion draft, June 2026). Its sponsors say in their FAQ that it "would federalize leading frontier safety laws including California's SB-53, New York's RAISE Act, and Illinois' SB315." In a title scan of House and Senate bills in the 119th Congress, we found no bill under that name.
• H.R. 9925, the FRONTIER Act. Rep. Obernolte introduced this bill on July 23, 2026, and it was referred to committee. Sec. 9(b) says no state may "adopt or enforce any law, regulation, order, or other requirement that imposes new substantive obligations on artificial intelligence developers" in listed subject areas, which include frontier AI risk transparency and third-party auditing. Sec. 9(c) has exceptions. It is a bill, not a law, and our preemption tracker does not treat it as the same text as the draft.

EO 14409, signed June 2, 2026, calls for a "voluntary framework" for early government access to covered frontier models. It has no state-law provision, and Section 3(c) says that section does not authorize "a mandatory governmental licensing, preclearance, or permitting requirement."

In a CourtListener search, we did not find a docket challenging SB 53 or the RAISE Act. A web search found no challenge to SB 315. Those are searches on limited channels, not a statement that no one has challenged these laws. Our preemption tracker follows the order, the bills and the cases.

What Changes in 2027

As of November 4, 2026, these are the dates on our calendar for frontier AI law. Any of them could change.

Pending matters to watch: H.R. 9925 in committee, the Great American AI Act draft, and Michigan's HB 4668, each at the status given above. New York's office may also adopt rules under Section 1429.

Change Log

• November 4, 2026: Page published. Status checked through the sources and dates named on this page.

Related AI Law Pages

• AI Laws in the US: State-by-State Tracker
• Can States Still Regulate AI? Preemption Tracker
• California AI Laws
• New York AI Laws
• Illinois AI Laws

Frequently Asked Questions

Which states have frontier AI safety laws?

As of November 4, 2026, by our count of enacted laws, and subject to the search limits shown on our AI laws hub, three states have frontier-model safety statutes. California's SB 53 is in force. New York's RAISE Act and Illinois SB 315 are scheduled to take effect January 1, 2027. Connecticut has a frontier whistleblower provision only.

What is California SB 53?

SB 53 is the Transparency in Frontier Artificial Intelligence Act, in force since January 1, 2026. Developers of frontier models trained with more than 10^26 operations must publish transparency reports and report critical safety incidents. Those with more than $500 million in annual gross revenues, with affiliates, must also publish a frontier AI framework.

What is the RAISE Act?

The RAISE Act is New York's frontier AI safety law. S8828, signed March 27, 2026, repealed the 2025 text and added a new article 44-B of the General Business Law, which takes effect January 1, 2027.

Is the RAISE Act in effect?

Not as of November 4, 2026. It is signed and scheduled to take effect January 1, 2027.

What is Illinois SB 315?

SB 315 is the Artificial Intelligence Safety Measures Act, P.A. 104-0538, approved July 6, 2026. It takes effect January 1, 2027. Every frontier developer owes 72-hour incident reports from that date. Large frontier developers owe a disclosure statement from that date, and a published framework and an annual third-party audit from 2028, with the audit starting on the later of January 1, 2028, or 90 days after first qualifying.

What is the difference between SB 53 and the RAISE Act?

The thresholds match. SB 53 is in force, gives 15 days to report a critical safety incident, and caps the penalty at $1 million per violation. The RAISE Act starts January 1, 2027, gives 72 hours, allows up to $1 million for a first violation and $3 million for later ones, and adds a disclosure statement and fee for large frontier developers, with a separate $1,000-a-day filing penalty. SB 53 has a whistleblower section; article 44-B does not.

What is a frontier AI model under state law?

In California, New York and Illinois, a frontier model is a foundation model trained using more than 10^26 integer or floating-point operations, counting later fine-tuning, reinforcement learning or other material modifications the developer applies to a preceding foundation model.

Do frontier AI laws require a third-party audit?

Illinois is the only one of the three that requires an annual third-party audit. It applies to large frontier developers, starts no earlier than January 1, 2028, and tests compliance with Section 10 of the act.

Can I sue a company under SB 53 or the RAISE Act?

The safety chapters do not give the public a right to sue. The Attorney General enforces the main civil penalties, and New York's office can separately impose $1,000 a day for failing to file a disclosure statement or correct false information. California's Labor Code 1107.1 separately gives covered employees a civil action over whistleblower violations. New York's article 44-B says it creates no private right of action, and Labor Law 740, the general whistleblower law, is separate.

Did Trump's executive order cancel SB 53 or the RAISE Act?

EO 14365 does not itself declare any state law preempted. As of November 4, 2026, we found no federal statute in the public laws of the 119th Congress that generally preempts or pauses state AI laws. Proposals in Congress, including H.R. 9925, are not law.