California Sets Precedent with AI Transparency Legislation
Sacramento, CA – September 19, 2024 – California Governor Gavin Newsom has enacted a pioneering piece of legislation known as the California AI Transparency Act, aimed at regulating the disclosure practices of generative artificial intelligence (AI) systems. This landmark law, the most detailed AI watermarking legislation in the United States, will come into effect on January 1, 2026.
Key Provisions and Requirements
The California AI Transparency Act mandates that entities, referred to as "covered providers," which develop generative AI systems must adhere to stringent disclosure policies. The term "covered providers" applies to those whose generative AI systems attract more than 1,000,000 monthly users or visitors and are accessible to the public within the state. Such systems range from those generating text to images, videos, and audio that emulate their underlying training data.
The Act introduces several crucial requirements for these covered providers:
AI Detection Tools: Providers must offer users AI detection tools at no charge. These tools are designed to help users verify whether the content has been created or altered using an AI system. They will also include system provenance data to verify the authenticity and origin of the content.
Content Disclosure Options: Providers must offer options for both "manifest" and "latent" disclosures:
Manifest Disclosures: These are clear and conspicuous notices identifying content as AI-generated, ensuring immediate recognition by users.
Latent Disclosures: Metadata embedded in the AI-generated content must contain information such as the provider's name, the AI system's name and version, content creation or alteration timestamps, and a unique identifier. These disclosures must be difficult to remove and detectable by AI detection tools.
Contractual Obligations: Providers licensing their generative AI systems to third parties must enforce these disclosure requirements through contractual agreements. Should providers become aware that a licensee is incapable of adhering to these requirements, they must revoke the license within 96 hours.
Implementation and Enforcement
Upon its commencement in 2026, compliance with this law will be monitored by the California Attorney General, city attorneys, and county counsels. Non-compliance may result in civil penalties amounting to $5,000 per day.
Context and Implications
California joins the states of Colorado, Utah, and Illinois in adopting AI transparency laws. However, the California AI Transparency Act stands out for its specificity regarding AI watermarking. This groundbreaking legislation seeks to address the increasing use and proliferation of generative AI technologies by ensuring transparency and accountability.
As companies in the AI industry look to meet these new regulations, they are likely to invest in developing and refining their generative AI systems to comply with these legal requirements. This Act also prompts licensors and licensees to review and potentially revise their agreements to include the necessary contractual stipulations mandated by the new law.
The adoption of this law marks a significant step in the regulatory landscape for AI technology, setting a comprehensive framework that could influence national and global standards in the future.
Source: Noah Wire Services