# What does California SB 53, the Transparency in Frontier Artificial Intelligence Act, require?

Source: https://superintelligencelicensing.com/guides/california-sb-53-requirements  
Publisher: ZSKFL Management  
Last updated: 2026-11-02. Not legal advice.

> California SB 53, the Transparency in Frontier AI Act, in force since January 1, 2026: who it covers, what each tier publishes, the 15-day incident clock.

## At a glance

- California’s SB 53, the Transparency in Frontier Artificial Intelligence Act (Chapter 138, Statutes of 2025), was signed September 29, 2025 and has been in effect since January 1, 2026.
- It covers “frontier developers” that trained a foundation model with more than 10²⁶ operations. All of them publish per-model transparency reports and report critical safety incidents within 15 days; “large frontier developers,” with more than $500 million in annual gross revenue counted with affiliates, also publish a frontier AI framework.
- Only the Attorney General can seek the civil penalty, up to $1 million per violation by a large frontier developer; covered employees can sue over retaliation under the Labor Code. There is no license, no pre-approval and no new liability rule. As of November 2, 2026, Anthropic and OpenAI have published frameworks they tie to the act, Microsoft’s framework names it, and the current xAI edition no longer does.

## What does California SB 53, the Transparency in Frontier Artificial Intelligence Act, require?

It requires the companies that train the largest AI models to publish their safety documents and to tell the state when something goes seriously wrong: a transparency report with each new frontier model, incident reports within 15 days and, above $500 million in revenue, a frontier AI framework the developer must follow plus quarterly internal-use risk summaries. The act also protects covered employees who report a danger from catastrophic risk or a violation of the act, and asks a consortium to design a public computing cluster called CalCompute.

Nothing in it lets a California agency approve, refuse or recall a model. The [frontier AI laws guide](https://superintelligencelicensing.com/guides/frontier-ai-laws) compares it with New York, Illinois and the EU.

## When did SB 53 take effect, and where does it sit in the code?

January 1, 2026. Senator Scott Wiener introduced the bill on January 7, 2025. The Assembly passed it 59 to 7 on September 12, 2025, the Senate concurred 29 to 8 the next day, and the Governor approved it on September 29, 2025. The chapter sets no effective date of its own, so article IV, section 8(c)(1) of the California Constitution applies: a statute enacted at a regular session takes effect on the January 1 after a 90-day period from enactment.

The chapter adds Business and Professions Code sections 22757.10 through 22757.16 (section 2), Government Code section 11546.8 (section 3) and Labor Code sections 1107 through 1107.2 (section 4); uncodified section 5 holds the severability, federal-conflict and local-preemption clauses. References below are to the Business and Professions Code unless stated.

## Who is a frontier developer, and who is a large frontier developer?

Section 22757.11 sets three nested definitions.

- **Frontier model** (subdivision (i)): “a foundation model that was trained using a quantity of computing power greater than 10^26 integer or floating-point operations,” counting “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** (subdivision (h)): “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” as the 10²⁶ line.
- **Large frontier developer** (subdivision (j)): a frontier developer that “together with its affiliates collectively had annual gross revenues in excess of five hundred million dollars ($500,000,000) in the preceding calendar year.”
Subdivision (h) turns on the computing power the person “has used, or intends to use.” The definition reaches a person who has “initiated the training of” a frontier model, not only one who has finished training it. Subdivision (i)(2) counts post-training compute; the text does not separately address a company that fine-tunes a model another developer trained. The [compute calculator](https://superintelligencelicensing.com/compute-calculator) estimates where a run lands, and [compute thresholds](https://superintelligencelicensing.com/guides/compute-thresholds) explains the counting. Unlike New York’s law, the text has no territorial clause and no university exemption.

**Catastrophic risk** (subdivision (c)) is a foreseeable and material risk that a frontier developer’s development, storage, use or deployment of a frontier model “will materially contribute to the death of, or serious injury to, more than 50 people or more than one billion dollars ($1,000,000,000) in damage to, or loss of, property arising from a single incident” in which the model gives expert-level help with a chemical, biological, radiological or nuclear weapon; engages, with no meaningful human oversight, in a cyberattack or in conduct that would be murder, assault, extortion or theft if a human did it; or evades the control of its developer or user. Information publicly accessible elsewhere, lawful federal activity and harm the model did not materially contribute to are excluded, and a loss of equity value is not property damage (section 22757.16).

**Critical safety incident** (subdivision (d)) is any of four events: unauthorized access to, modification of or exfiltration of model weights that results in death or bodily injury; harm from a catastrophic risk materializing; loss of control causing death or bodily injury; or a frontier model that “uses deceptive techniques against the frontier developer to subvert the controls or monitoring of its frontier developer outside of the context of an evaluation designed to elicit this behavior and in a manner that demonstrates materially increased catastrophic risk.” The fourth prong has no death-or-injury trigger: deception of that kind, outside an evaluation and in a manner that demonstrates materially increased catastrophic risk, is reportable without any harm.

## What must a large frontier developer publish?

A frontier AI framework. Section 22757.12(a) requires a large frontier developer to “write, implement, comply with, and clearly and conspicuously publish on its internet website” a framework describing how it approaches ten things:

- national standards, international standards and industry-consensus best practices;
- the thresholds, which may be tiered, that identify whether a model could pose a catastrophic risk;
- mitigations;
- reviewing assessments and the adequacy of mitigations as part of the decision to deploy a model or use it extensively internally;
- third-party assessment;
- revisiting the framework, including when a model is modified enough to need a new transparency report;
- cybersecurity for unreleased model weights;
- identifying and responding to critical safety incidents;
- internal governance;
- catastrophic risk from internal use, “including risks resulting from a frontier model circumventing oversight mechanisms.”
Subdivision (a) ties the framework to no model release or other date. Once written, it must be reviewed, and updated as appropriate, at least once a year, and a material modification must be published with a justification within 30 days (subdivision (b)). Every three months, or on another reasonable schedule stated in writing, the developer must send the Office of Emergency Services “a summary of any assessment of catastrophic risk resulting from internal use of its frontier models” (subdivision (d)).

The words “comply with” matter. Section 22757.15 lists failing to comply with the framework alongside failing to publish a required document, under the same cap of up to $1 million per violation, with the amount depending on severity. The company writes its own framework, but once it is posted, the Attorney General can measure the company against it.

## What must every frontier developer publish?

A transparency report, before or at the same time as deploying a new or substantially modified frontier model to a third party. Section 22757.11(e) defines “deploy” as making a frontier model “available to a third party for use, modification, copying, or combination with other software,” excluding availability to a third party “for the primary purpose of developing or evaluating the frontier model.” For a large frontier developer, catastrophic risk from internal use goes into the quarterly summaries under section 22757.12(d).

Section 22757.12(c)(1) lists seven items: the developer’s website; a way for a natural person to contact it; the model’s release date; supported languages; supported output modalities; intended uses; and any generally applicable restrictions on use. A large frontier developer adds summaries of its catastrophic-risk assessments under the framework, their results, the extent of third-party evaluator involvement and the other steps it took under the framework (paragraph (2)). Publishing this “as part of a larger document, including a system card or model card, shall be deemed in compliance” (paragraph (3)).

Subdivision (e) bars any frontier developer from materially false or misleading statements about catastrophic risk from its frontier models or its management of catastrophic risk, and a large frontier developer from such statements about its implementation of, or compliance with, its framework; it does not apply to a statement made in good faith that was reasonable under the circumstances. Subdivision (f) allows redactions for trade secrets, cybersecurity, public safety, national security or compliance with law, if the developer describes “the character and justification of the redaction” and keeps the unredacted version for five years.

**SB 53 duties by tier, Business and Professions Code sections 22757.12 to 22757.15 and Labor Code section 1107.1**

| Duty | Frontier developer | Large frontier developer |
|---|---|---|
| Frontier AI framework, published and followed; annual review; changes within 30 days | No | Yes (22757.12(a), (b)) |
| Transparency report per model: basic items | Yes (22757.12(c)(1)) | Yes |
| Transparency report: risk-assessment summaries, third-party involvement | No | Yes (22757.12(c)(2)) |
| Internal-use risk summary to the Office of Emergency Services every three months | No | Yes (22757.12(d)) |
| Incident report within 15 days; 24-hour disclosure if imminent risk | Yes (22757.13(c)) | Yes |
| No false or misleading statements about catastrophic risk | Yes (22757.12(e)(1)(A)) | Yes, plus framework compliance (22757.12(e)(1)(B)) |
| No rule, policy or contract that blocks or punishes protected disclosures; notice to covered employees of their rights | Yes (Labor Code 1107.1(a), (b), (d)) | Yes |
| Anonymous whistleblower channel with monthly updates | No | Yes (Labor Code 1107.1(e)) |
| Civil penalty up to $1 million per violation | Not named in 22757.15 | Yes (22757.15) |

## How are critical safety incidents reported, and how fast?

Section 22757.13 sets two clocks. A frontier developer “shall report any critical safety incident pertaining to one or more of its frontier models to the Office of Emergency Services within 15 days of discovering the critical safety incident” (subdivision (c)(1)), and may amend the report later. If the incident “poses an imminent risk of death or serious physical injury,” the developer “shall disclose that incident within 24 hours to an authority, including any law enforcement agency or public safety agency with jurisdiction, that is appropriate based on the nature of that incident and as required by law” (subdivision (c)(2)). Paragraph (1) opens “Subject to paragraph (2),” and paragraph (2) names “an authority,” not the Office of Emergency Services; the text does not say whether a 24-hour disclosure also satisfies the 15-day report. Neither paragraph is limited to deployed models: the duty covers incidents “pertaining to one or more of its frontier models.”

The office must run a reporting mechanism for frontier developers and the public; a report gives the incident date, why it qualifies, a short and plain description, and whether internal use was involved (subdivision (a)). Its page for the act, run by the California Cybersecurity Integration Center, lists a secure “TFAIA Portal” for incident reports and internal-use summaries, with a downloadable worksheet of the portal’s fields. The page carries no publication date.

Reports are exempt from the California Public Records Act (subdivision (f)). Beginning January 1, 2027, the office must produce an annual report with anonymized and aggregated information on the incidents it has reviewed and send it to the Legislature and the Governor (subdivision (g)). Subdivisions (h) through (j) add a federal off-ramp for incident reporting only: the office may designate by regulation a federal law, regulation or guidance document with a substantially equivalent or stricter standard, and a developer that declares it will comply that way is deemed in compliance to the extent it does. We found no such designation as of November 2026.

## What protections do whistleblowers get?

Labor Code sections 1107 to 1107.2 protect “covered employees,” meaning employees “responsible for assessing, managing, or addressing risk of critical safety incidents.” A frontier developer may not adopt a rule, policy or contract that prevents a covered employee from disclosing, or retaliate for disclosing, information to the Attorney General, a federal authority, a manager or another covered employee with authority to investigate. The protection applies where the employee has reasonable cause to believe the developer’s activities “pose a specific and substantial danger to the public health or safety resulting from a catastrophic risk” or that it has violated the act (section 1107.1(a)). Every frontier developer must give covered employees clear notice of their rights and responsibilities, including by a workplace posting or by annual written notice that employees acknowledge (1107.1(d)).

A large frontier developer must also provide “a reasonable internal process through which a covered employee may anonymously disclose information,” with “a monthly update to the person who made the disclosure” (1107.1(e)). Once an employee shows that protected activity contributed to an adverse action, the developer must prove “by clear and convincing evidence” that it would have acted anyway (1107.1(g)); attorney’s fees and injunctive relief are available.

The Labor Code definitions reach further than the Business and Professions Code ones. Section 1107 keys on a “foundation model,” not a “frontier model,” and its weight-theft prong covers “death, bodily injury, or damage to, or loss of, property,” where section 22757.11(d)(1) stops at death or bodily injury. The wider definition shapes who counts as a covered employee (section 1107(b)). Protection under section 1107.1(a) still turns on what the employee reasonably believes the information shows: a specific and substantial danger resulting from a catastrophic risk, or a violation of the act.

## Who enforces SB 53, and what are the penalties?

Only the Attorney General can recover the civil penalty; covered employees bring their own retaliation suits under Labor Code section 1107.1. Under section 22757.15, a large frontier developer that “fails to publish or transmit a compliant document,” makes a statement in violation of section 22757.12(e), “fails to report an incident as required by Section 22757.13, or fails to comply with its own frontier AI framework shall be subject to a civil penalty in an amount dependent upon the severity of the violation that does not exceed one million dollars ($1,000,000) per violation,” recovered “in a civil action brought only by the Attorney General.”

That sentence names large frontier developers and no one else. A frontier developer under the $500 million line still carries the transparency-report, incident-report and no-false-statement duties and the Labor Code whistleblower duties, but section 22757.15 sets no dollar figure for its failures, and section 5(c) of the chapter keeps every other duty and remedy in place.

As of November 2026 we found no Attorney General enforcement action under the act, only private allegations. The Midas Project, a watchdog group, said in February 2026 that OpenAI released GPT-5.3-Codex without the safeguards its own Preparedness Framework required, and in September 2026 that OpenAI had not assigned the risk tiers its Frontier Governance Framework calls for, in any category, to GPT-5.6 preview, GPT-5.6 and GPT-6 Astra. OpenAI told Fortune both times that it was confident in its compliance with SB 53; the Attorney General’s office told Fortune in February that it could not confirm or deny any investigation.

## Who can change the thresholds, and when?

The Legislature does, by amending the statute; the Department of Technology only recommends. Section 22757.14 requires the department, “on or before January 1, 2027, and annually thereafter,” to assess recent evidence and recommend whether and how to update the three coverage definitions, weighing similar thresholds in international standards or federal law, stakeholder input, whether a developer can tell before training or deployment whether it is covered, how complex that determination is, and whether coverage is verifiable by outsiders.

Assembly versions of the bill from July 8 to September 2, 2025 let the Attorney General adopt regulations updating the definition of a large developer. The September 5 amendments replaced that with the department’s recommendations, and the chaptered text gives no agency rulemaking power over the definitions or the documents. We found no 2026 amendment. OpenAI said in August 2026 that the act should be amended to add monitoring of frontier models during training and evaluation; we found no bill carrying that request.

## What is CalCompute?

On paper, a consortium and a report. Government Code section 11546.8 creates a 14-member consortium within the Government Operations Agency to develop a framework for “a public cloud computing cluster to be known as CalCompute,” one that fosters research that benefits the public and expands access to computational resources. The consortium must make reasonable efforts to establish CalCompute within the University of California to the extent possible.

On or before January 1, 2027, the agency must submit the consortium’s report to the Legislature, covering existing cloud infrastructure, cost and funding, governance, parameters for use, workforce, proposed partnerships and how CalCompute can prioritize the use of the current public sector workforce; the consortium then dissolves. Subdivision (k) is the gate: the section “shall become operative only upon an appropriation in a budget act, or other measure, for the purposes of this section.” We found no public announcement, as of November 2026, that the consortium has been appointed.

## What does SB 53 not do?

- **No license and no pre-approval.** No section requires permission before training, deploying or releasing a model. [AI licensing explained](https://superintelligencelicensing.com/guides/ai-licensing-explained) sets out the difference between a permit and a transparency duty.
- **No registration and no fee.** There is no disclosure statement, beneficial-owner filing or regulator assessment of the kind in New York’s law.
- **No liability regime.** Beyond the Attorney General’s civil penalty and the employee remedies in Labor Code section 1107.1, the act neither makes a developer liable for harm a model causes nor shields one from existing remedies (section 5(c)).
- **No audit.** The framework must describe third-party assessment and the transparency report must summarize it, but no independent audit is required.
- **Local rules are displaced.** Section 5(f) preempts any local rule adopted on or after January 1, 2025 “specifically related to the regulation of frontier developers with respect to their management of catastrophic risk.”

## How does SB 53 interact with the federal preemption push?

As of November 2026, it remains in force and no court has enjoined it. Executive Order 14365 of December 11, 2025 directed the Attorney General to establish an AI Litigation Task Force within 30 days to challenge state AI laws; the Attorney General did so by memorandum on January 9, 2026, and on April 24, 2026 the Justice Department moved to intervene in xAI’s lawsuit against Colorado’s AI Act. We found no federal suit against SB 53. xAI’s December 2025 suit, X.AI LLC v. Bonta, challenges AB 2013, the training-data disclosure law, not SB 53.

Section 5(e) of the chapter says the act “shall not apply to the extent that it is preempted by federal law.” The clause names no federal law, and an executive order cannot by itself void a state statute. Executive Order 14365 directs the Federal Communications Commission to begin a proceeding on whether to adopt a federal reporting and disclosure standard that preempts conflicting state laws (section 6) and calls for a legislative recommendation (section 8). The task force, the FRONTIER Act and the letter from 26 state attorneys general are tracked in the [frontier AI laws guide](https://superintelligencelicensing.com/guides/frontier-ai-laws#is-the-u-s-federal-government-overriding-state-laws), and the federal orders in the [AI executive orders list](https://superintelligencelicensing.com/guides/ai-executive-orders-list).

## Which developers have published SB 53 documents?

As of November 2, 2026, Anthropic and OpenAI each identify a published framework as the one they use for the act. Microsoft’s framework names the act without saying whether Microsoft is covered, and xAI’s current edition no longer mentions it. We did not verify any company’s revenue or training compute, and the act does not require anyone to declare coverage.

**Published frontier AI frameworks and the Transparency in Frontier Artificial Intelligence Act, as of November 2026**

| Developer | Document | Date | What it says about the act |
|---|---|---|---|
| Anthropic | Frontier Compliance Framework | December 19, 2025 | Will serve as its compliance framework for SB 53; the Responsible Scaling Policy remains voluntary. |
| xAI | Frontier Artificial Intelligence Framework | December 30, 2025; new edition effective June 30, 2026 | The December 2025 edition said it “complies with California’s Transparency in Frontier Artificial Intelligence Act,” citing section 22757.10 et seq. The June 30, 2026 edition does not mention the act. |
| Microsoft | Frontier Governance Framework | February 2026 | Names “California’s Transparency in Frontier AI Act (TFAIA)” among the laws that put a model in scope for its leading-indicator assessment, and lists the act in its change log as a reason for the update; does not say whether Microsoft is a frontier developer. |
| OpenAI | Frontier Governance Framework | May 28, 2026 | “Under California’s Transparency in Frontier AI Act (TFAIA), this FGF is our Frontier AI Framework”; the Preparedness Framework remains in use alongside it. |
| Google DeepMind | Frontier Safety Framework, version 3.1 | April 17, 2026 | Does not name the act; per-model reports exist for Gemini 3 Pro and Gemini 3.7 Flash. |
| Meta | Meta Superintelligence Scaling Framework, version 2.1 (formerly Frontier AI Framework) | October 2026 | Does not name the act in the text we read. |

None of these documents is a transparency report. That duty is per model, and whether a given system card carries the section 22757.12(c) items is a question about that card.

This guide states the chaptered text of SB 53 as of November 2, 2026 and is general information, not legal advice.

## Sources

- [SB 53, Chapter 138, Statutes of 2025, chaptered text](https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53), California Legislative Information: Business and Professions Code §§ 22757.10–22757.16, Government Code § 11546.8, Labor Code §§ 1107–1107.2, and uncodified sections 1, 5 and 6.
- [SB 53 bill history](https://leginfo.legislature.ca.gov/faces/billHistoryClient.xhtml?bill_id=202520260SB53), California Legislative Information: introduced January 7, 2025; Assembly passage September 12, 2025 (59–7); Senate concurrence September 13, 2025 (29–8); approved by the Governor and chaptered September 29, 2025.
- [SB 53 bill page and text versions](https://calmatters.digitaldemocracy.org/bills/ca_202520260sb53), CalMatters Digital Democracy: Assembly amendments of July 8, July 17, September 2 and September 5, 2025.
- [California Constitution, article IV, section 8](https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CONS§ionNum=SEC.%208.&article=IV), California Legislative Information (subdivision (c)(1), effective date of statutes enacted at a regular session).
- [Governor Newsom signs SB 53, advancing California’s world-leading artificial intelligence industry](https://www.gov.ca.gov/2025/09/29/governor-newsom-signs-sb-53-advancing-californias-world-leading-artificial-intelligence-industry/), Office of Governor Gavin Newsom, September 29, 2025.
- [Transparency in Frontier AI Act reporting](https://www.caloes.ca.gov/office-of-the-director/operations/homeland-security/california-cybersecurity-integration-center/transparency-in-frontier-ai-act-reporting/), California Governor’s Office of Emergency Services, California Cybersecurity Integration Center, accessed November 2, 2026.
- [Sharing our compliance framework for California’s Transparency in Frontier AI Act](https://www.anthropic.com/news/compliance-framework-SB53), Anthropic, December 19, 2025.
- [xAI Frontier Artificial Intelligence Framework](https://data.x.ai/2025-12-31-xai-frontier-artificial-intelligence-framework.pdf), xAI, last updated December 30, 2025 (PDF), and the [edition effective June 30, 2026](https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-f1f0c916.pdf) (PDF).
- [Frontier Governance Framework](https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Frontier-Governance-Framework-Feb-2026.pdf), Microsoft, February 2026 (PDF).
- [OpenAI’s Frontier Governance Framework](https://openai.com/index/openai-frontier-governance-framework/), OpenAI, May 28, 2026, and the [framework PDF](https://cdn.openai.com/pdf/e37d949b-8c9f-4d76-b99e-4272f4631a7e/openai-frontier-governance-framework.pdf).
- [Frontier Safety Framework](https://deepmind.google/frontier-safety/), Google DeepMind, version list and per-model reports, and the [version 3.1 PDF](https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf), published April 17, 2026.
- [Meta Superintelligence Scaling Framework, version 2.1](https://research.meta.ai/static/meta-superintelligence-scaling-framework), Meta, October 2026 (PDF).
- [OpenAI disputes watchdog allegation it violated California’s new AI law with GPT-5.3-Codex release](https://fortune.com/2026/02/10/openai-violated-californias-ai-safety-law-gpt-5-3-codex-ai-model-watchdog-claims/), Fortune, February 10, 2026.
- [OpenAI may have violated California’s AI safety law with latest model releases, AI watchdog says](https://fortune.com/2026/09/14/openai-may-have-violated-californias-ai-safety-law-with-astra-model-releases-ai-watchdog-says/), Fortune, September 14, 2026.
- [OpenAI calls for California to strengthen its AI safety laws](https://engadget.com/2242200/openai-calls-for-california-to-strengthen-ai-safety-laws), Engadget, August 22, 2026.
- [Ensuring a National Policy Framework for Artificial Intelligence](https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy/) (Executive Order 14365, 90 FR 58499), The White House, December 11, 2025.
- [Artificial Intelligence Litigation Task Force](https://www.justice.gov/ag/media/1422986/dl), memorandum from the Attorney General to all Department of Justice employees, January 9, 2026 (PDF).
- [Justice Department Intervenes in xAI lawsuit Challenging Colorado’s ‘Algorithmic Discrimination’ Law](https://www.justice.gov/opa/pr/justice-department-intervenes-xai-lawsuit-challenging-colorados-algorithmic-discrimination), U.S. Department of Justice, Office of Public Affairs, April 24, 2026.
- [X.AI LLC v. Rob Bonta, No. 2:25-cv-12295 (C.D. Cal.)](https://www.courtlistener.com/docket/72086154/xai-llc-v-rob-bonta/), docket, CourtListener.

## Common questions

**Who has to comply with California SB 53?**

Frontier developers: anyone who trained, or initiated training of, a foundation model using more than 10^26 integer or floating-point operations, counting fine-tuning, reinforcement learning and other material modifications. Those with more than $500 million in annual gross revenue in the preceding calendar year, counted with affiliates, are large frontier developers and must also publish a frontier AI framework, add risk-assessment summaries to each transparency report, send quarterly internal-use risk summaries to the Office of Emergency Services and run an anonymous whistleblower channel.

**How fast must a critical safety incident be reported under SB 53?**

Within 15 days of discovery, to the California Office of Emergency Services, through the TFAIA Portal the office runs. If the incident poses an imminent risk of death or serious physical injury, the developer must disclose it within 24 hours to an appropriate authority, which can include a law enforcement or public safety agency with jurisdiction. Beginning January 1, 2027, the office must produce an annual report with anonymized and aggregated information on the incidents it has reviewed and send it to the Legislature and the Governor.

**Does SB 53 require a license or government approval for AI models?**

No. SB 53 is a transparency and reporting statute. It requires published frameworks and transparency reports, incident reports and whistleblower protections. The Attorney General can seek civil penalties of up to $1 million per violation from large frontier developers, and covered employees can sue over retaliation. No agency may approve, refuse or recall a model, there is no registration or fee, and the act creates no new liability for harm a model causes.