Introduction
On September 19, 2024 [2], California Governor Gavin Newsom enacted the California Artificial Intelligence Transparency Act (CAITA) by signing SB 942 [2]. This legislation [2], effective January 1, 2026 [1] [2], is designed to enhance transparency in the use of artificial intelligence (AI) systems [2], particularly in the creation of synthetic and deepfake content [2]. It targets providers of generative AI systems with over one million monthly users that are publicly accessible in California [1].
Description
On September 19, 2024 [2], California Governor Gavin Newsom signed SB 942 [2], known as the California Artificial Intelligence Transparency Act (CAITA) [2], which will take effect on January 1, 2026 [1] [2]. This legislation aims to enhance transparency in the deployment of artificial intelligence (AI) systems [2], particularly regarding the creation of synthetic and deepfake content [2]. CAITA applies to providers of generative AI systems that are publicly accessible in California and have over one million monthly users, including companies that license and integrate such technology [2].
The law defines a generative AI system as one capable of producing derived synthetic content [2], including text [2], images [2], video [2], and audio [2], based on its training data [2]. It specifies that AI refers to engineered systems that can generate outputs influencing environments [2]. Covered Providers are mandated to implement AI detection and disclosure features on their websites and mobile applications [2], offering users a no-cost AI Detection Tool. This tool allows users to upload content or input a URL to assess whether it was created or altered by a generative AI system [2]. It must be publicly accessible [2], support an API [2], and output Provenance Data without identifying users or their devices [2].
Provenance Data is defined as information embedded in digital content to verify its authenticity and modification history [2]. Covered Providers and third-party licensees are required to maintain these disclosures and tools without modifications that would remove them [2]. Providers are prohibited from retaining submitted content longer than necessary and from collecting personal information [2], except for feedback submissions [2]. Feedback on the AI detection tool’s efficacy must be collected and utilized for improvements [2].
Exemptions exist for products or services that solely provide non-user-generated content in video games [2], television [2], streaming [2], movies [2], or interactive experiences [2]. Notably, third-party licensees outside California may still be subject to CAITA if their generative AI systems are accessible from within the state [2].
Noncompliance with CAITA can result in significant penalties [1], including civil fines of $5,000 for each day of violation [1], with each day considered a separate offense [2]. Third-party licensees may also face contractual breaches and civil actions from the California Attorney General and local prosecutors [2]. Compliance with CAITA may require substantial effort [2], and industry standards may evolve following its enforcement. Recent examples of AI detection tools include the C2PA Specification [2], which binds Provenance Data to media [2], and detection classifiers will evaluate the likelihood that content was generated by a generative AI system [2]. While there is no requirement for latent disclosures in synthetic text [2], third-party licensees must maintain the capability to include such disclosures if their licensed generative AI system allows it [2]. If a license is revoked due to modifications affecting disclosure capabilities [2], third-party licensees must cease using the system [2].
Conclusion
The enactment of CAITA represents a significant step towards ensuring transparency in the use of AI technologies, particularly in the realm of synthetic content creation. By mandating detection and disclosure tools, the legislation aims to protect consumers and maintain the integrity of digital content. The law’s stringent compliance requirements and potential penalties underscore the importance of adherence, which may drive industry standards and practices. As a result, companies operating within California or offering services accessible from the state must prepare for the operational and legal implications of this legislation.
References
[1] https://www.jdsupra.com/legalnews/california-s-ai-transparency-act-1374224/
[2] https://www.gunder.com/en/news-insights/insights/client-insight-california-ai-transparency-act
