2025/11/2
California Governor Newsom Signs Landmark "Transparency in Frontier AI Act" (SB 53,TFAIA)

Governor Gavin Newsom of California signed the landmark "Transparency in Frontier AI Act" (SB 53) on September 29, 2025, aimed at enhancing transparency in the development of frontier AI models by corporations.


Key Highlights

• Targeted Scope: The legislation is narrowly focused, regulating only large corporations with annual revenue exceeding $500 million that develop models trained with over 10²⁶ floating-point operations (FLOPS). This scope is designed to avoid burdening small and medium-sized enterprises (SMEs) and to protect the innovation ecosystem. Companies like Anthropic and OpenAI are among those targeted.
• Four Core Obligations for Transparent Governance: Publication of Frontier AI Framework, Publication of Transparency Report, Reporting of Critical Safety Incidents, and Whistleblower Protections. Violations could result in fines up to $1 million.
• Dynamic Adjustment Mechanism: The California Department of Technology will annually evaluate and update the definitions of "frontier model" and "developer" to ensure the legislation remains current with technological advancements. The act also establishes the public cloud computing platform, CalCompute, to support academic research and development.

The bill was introduced by Democratic Senator Scott Wiener, who also proposed the earlier, Newsom-vetoed "AI Regulation Act" (SB 1047). SB 53 is considered a revised and adjusted version of SB 1047.
The passage of this law not only fills a legislative void in the United States concerning frontier AI models but is also expected to have a significant impact on AI governance across the US and globally, given that California is home to the headquarters of many of the world's leading AI companies.
Scope of Regulation
SB 53 regulates explicitly "Large Frontier Developers" (defined as companies with over $500 million in revenue in the prior year) and the "Frontier Models" they train (defined as models trained with a computational load exceeding10²⁶ FLOPS).
The legislation's purpose is to prevent the catastrophic risks potentially posed by these high-level models while avoiding the imposition of undue burdens on SMEs or lower-scale AI systems, thus maintaining innovation vitality.
Consequently, SB 53 is essentially designed for large California-based AI developers such as Anthropic, OpenAI, Meta, and Google.

SB53 includes four core obligations:
 

1. Publication of Frontier AI Framework

 Large frontier developers must publish a “Frontier AI Framework” on their website, describing: (1) how they incorporate national, international and best practices; (2) governance structures for managing AI safety; and (3) documented procedures for identifying and mitigating “catastrophic risk”—defined as foreseeable, material risk that a frontier model will cause the death of, or serious injury to, more than 50 people, or cause over $1 billion in property damage, in a single incident, among other requirements. 

2. Publication of Transparency Report

Before or during deployment of a frontier model or a substantially modified version of such model, a frontier developer must publish on its website a transparency report. Reports must include frontier model details such as the release date, intended uses, conditions, restrictions, and languages supported by the frontier model. In addition to the above, large frontier developers’ transparency reports must also include summaries of catastrophic risk assessments from the frontier model, their results, and the role of any third-party evaluators.

3. Reporting of Critical Safety Incidents

All frontier developers are required to disclose critical safety incidents to the OES within 15 days of discovering the incident. If the critical safety incident “poses an imminent risk of death or serious physical injury,” the frontier developer is required to disclose such incident to a governmental authority (e.g., law enforcement agency or public safety agency) within 24 hours. 

4. Whistleblower Protections

The TFAIA also offers protections to employees who report potential catastrophic risks to authorities. Accordingly, employers are banned from retaliating against employees who report any risks they uncover while working for frontier developers. Employees can report frontier developer activities that pose a specific and substantial danger to public health or safety, or violate the TFAIA, if they have reasonable cause to believe such misconduct has occurred. 

To ensure safe and sustainable AI use, the TFAIA grants the California Attorney General broad authority to enforce violations of the law, with civil penalties of up to $1 million per violation, based on the severity of the offense. 

Additionally, the TFAIA also requires establishing CalCompute, a public computing cluster dedicated to advancing safe, ethical, and sustainable AI development for the public sector. Its primary goal is to foster research and innovation that directly benefits California’s residents, prioritizing public good over corporate interests.

Acknowledging that AI is rapidly advancing, the TFAIA requires the California Department of Technology to regularly review AI advancements and update the law to keep pace with the industry.

Following the passage of SB 53, there are three key points for future observation

The first point of observation is how SB 53 addresses and revises several highly debated provisions from its predecessor, SB 1047.
SB 1047 previously drew intense backlash from the industry because it mandated a "kill switch" for models in the event of a catastrophic incident. It also required developers to complete comprehensive safety protocols and undergo annual third-party audits before training, and mandated reporting of incidents within 72 hours. Violations were subject to high fines, calculated as a percentage of revenue (10% for the first offense, 30% for repeat offenses). In contrast, SB 53 removes the "kill switch" requirement. It shifts the focus to post-event risk disclosure and notification mechanisms. Furthermore, the penalties are standardized to a maximum of $1 million (approximately NTD $32 million) per violation, significantly reducing the cost of compliance for businesses and enhancing the feasibility of enforcing the act.

Secondly, SB 53 is anticipated to become a demonstrative regulatory framework for other U.S. states pushing for AI governance.Upon signing the bill, Governor Newsom explicitly stated that SB 53 will serve as a "blueprint for AI regulation in other states." This indicates California's intention to lead interstate policy coordination in the absence of a unified federal framework.
For instance, the Responsible AI Safety and Education (RAISE) Act (A6953) currently under discussion in New York State may reference the SB 53 framework. If passed, it could become the second state-level regulatory act targeting frontier AI in the U.S.
Finally, SB 53 may intensify the policy divergence between California and the Federal government.
Upon taking office, President Trump launched the "AI Action Plan," integrating over 90 federal AI policies. This plan is built upon three main pillars: "Encouraging Innovation," "Building AI Infrastructure," and "Developing AI Diplomacy and National Defense." Crucially, the plan requires the Office of Management and Budget (OMB) to review whether states are inhibiting AI development through excessive regulation. States found to be doing so could face cuts to federal subsidies.
Governor Newsom has previously clashed openly with President Trump on immigration issues. His current assertive stance in the field of AI governance is seen as part of his political strategy to secure the Democratic presidential nomination in 2028. It also demonstrates California's apparent attempt to seize a leadership position in AI regulatory strategy.