California Enacts Landmark AI Transparency Law: The Transparency in Frontier Artificial Intelligence Acth2>
On September 29, 2025, California Governor Gavin Newsom signed Senate Bill 53, known as the b>Transparency in Frontier Artificial Intelligence Actb> (“b>TFAIAb>”), establishing a comprehensive legal framework aimed at ensuring b>transparencyb>, b>safetyb>, and b>accountabilityb> in the development and deployment of advanced artificial intelligence (b>AIb>) models. This legislation marks California as the first state in the United States to implement such rigorous transparency measures for AI.p>
Scope of the TFAIAh3>
The TFAIA introduces new transparency and governance requirements specifically for organizations creating certain advanced AI systems, referred to as b>”frontier models.”b> Notably, the law distinguishes between different types of developers:p>
-
li>b>Frontier model:b> A foundation model trained using more than 10sup>26sup> integer or floating-point operations.li>
li>b>Frontier developer:b> A person who has trained or initiated the training of a frontier model.li>
li>b>Large frontier developer:b> A frontier developer with annual gross revenues exceeding US$500 million in the previous calendar year.li>
ul>
Furthermore, the California Department of Technology is empowered to update these definitions as technological advancements occur.p>
Key Obligations Under the TFAIAh3>
Frontier AI Frameworkh4>
Large frontier developers must create and publish a detailed b>Frontier AI Frameworkb> that outlines how they identify, assess, and mitigate catastrophic risks throughout the lifecycle of their models. This framework needs to be updated at least annually and within 30 days of any significant modifications. It should include:p>
-
li>Documentation of governance structures.li>
li>Mitigation processes.li>
li>Cybersecurity practices.li>
li>Alignment with national or international standards and industry best practices.li>
ul>
A b>catastrophic riskb> is defined as a foreseeable risk that could lead to significant harm, including:p>
-
li>Deaths or serious injuries to more than 50 individuals.li>
li>Property damage exceeding $1 billion.li>
ul>
Publication of Transparency Reportsh4>
Before deploying new frontier models, all developers must publish a b>transparency reportb>, which should include:p>
-
li>A communication mechanism for individuals to contact the developer.li>
li>The release date of the frontier model.li>
li>The modalities of outputs supported by the model.li>
li>Intended uses and any restrictions on those uses.li>
ul>
Large frontier developers have additional requirements, such as:p>
-
li>Catastrophic risk assessments.li>
li>Disclosure of third-party involvement in risk assessment.li>
li>Regular summaries of risk assessments related to internal use submitted to the California Office of Emergency Services.li>
ul>
Critical Safety Incident Reportingh4>
The TFAIA mandates that developers report any b>critical safety incidentsb> to the Office of Emergency Services within 15 days of discovery. This includes incidents that pose imminent risks of severe injury or death, with reports kept confidential to protect trade secrets.p>
Whistleblower Protectionsh4>
Whistleblower protections are established under the TFAIA, ensuring that employees can report significant health and safety risks without fear of retaliation. Developers must inform employees of their rights and provide systems for anonymous reporting.p>
Enforcement and Implementationh3>
The TFAIA empowers the California Attorney General to enforce compliance, imposing penalties of up to $1 million for violations. The law is set to take effect on January 1, 2026.p>
Conclusionh3>
The TFAIA represents a significant shift from voluntary industry standards to a mandatory legal framework for AI transparency. Governor Newsom has positioned this law as a potential model for other states, especially given the current lack of comprehensive federal AI regulations. While the national impact remains to be seen, the TFAIA may pave the way for a patchwork of state-level AI regulations similar to the California Consumer Privacy Act’s influence on privacy laws.p>
