California modifies AI transparency law weeks after it took effect: What SB 1000 changes and why it matters

Quick Refresher: What Was SB 942?

Signed by Governor Newsom in September 2024, SB 942 (the California AI Transparency Act, Bus. & Prof. Code §§ 22757 et seq.) created first-of-its-kind obligations for generative AI providers: embed provenance metadata in AI-generated content, offer free detection tools, and monitor licensees. The original law only applied to GenAI systems with over one million monthly users. 

Enter SB 1000: The “Fix-It” Bill That’s Actually an Expansion

SB 1000, which passed on August 27, 2026 as an urgency statute (meaning it takes effect immediately upon being signed by the Governor), doesn’t just fix typos. It fundamentally reshapes SB 942 in several important ways.

The user threshold is gone. The original law’s “over 1,000,000 monthly visitors or users” requirement has been deleted. Under SB 1000, a “covered provider” is now simply any person that creates, codes or otherwise produces a GenAI system that is publicly accessible within California. That’s a dramatic expansion — startups, mid-market SaaS companies and open-source projects with a public-facing interface all potentially fall within scope now.

The visible watermark requirement is gone; latent provenance data is all that remains. SB 942 had two separate disclosure requirements. First, a “manifest disclosure” obligation: covered providers had to offer users the option to slap a visible, human-readable label on AI-generated content (i.e. a watermark saying “AI-generated”). Second, a latent disclosure obligation: embedded provenance metadata baked into the file itself. SB 1000 eliminates the manifest disclosure requirement entirely.

Visible watermarks are no longer required – the statute now focuses exclusively on latent provenance data. Under SB1000, covered providers shall include a latent disclosure in any image, video or audio content created or altered by their GenAI system, which must convey the provider's name, system version information, a timestamp, a unique identifier, and whether the content was created or altered by AI.

New carve-out for “minor modifications.” SB 1000 introduces a defined list of changes that don’t trigger disclosure obligations: brightness/contrast adjustments, sharpening, saturation, file resizing, scaling, cropping, file format conversions and audio denoising. These lightweight edits are not “altering” content for purposes of the statute.

Assistive technology gets special treatment. SB 1000 adds a new definition of “assistive technology” (tools that increase, maintain or improve functional capabilities of individuals with disabilities) and requires that latent disclosures flag whether the GenAI system primarily functions as assistive technology. While SB 1000 does not exempt assistive technology from the disclosure requirements, the flag allows consumers and enforcement authorities to distinguish accessibility-focused tools from other GenAI systems — a contextual marker that may inform how the law is applied in practice.

The “AI detection tool” is now a “disclosure verification tool.” This development isn’t just a name change. The retooled provision allows covered providers to satisfy their verification obligations by directing users to a compliant third-party tool so long as it’s compatible with the provider’s latent disclosures and clearly accessible through the provider’s own interface. This change gives smaller providers a viable path to compliance without building detection infrastructure from scratch.

Stronger personal information protections. Covered providers are now prohibited from collecting, using, retaining, selling, sharing or otherwise making available personal information derived from users of the verification tool or from content processed by it, beyond what is strictly necessary to comply with the chapter. 

License enforcement: 72 hours, broader scope, no monitoring duty. SB 1000 shortens the license enforcement clock — from the original 96 hours under SB 942 down to 72 hours — after a covered provider discovers that an identifiable licensee has modified a system out of compliance. The covered provider can now revoke, suspend, or terminate the licensee's authorization (a broader menu of remedies than before). It also broadens the licensee’s cessation obligation; a terminated licensee must stop using and making the system available, including any copies or modified versions. At the same time, SB 1000 explicitly states that covered providers have no obligation to monitor, investigate or inquire into how licensees use or modify their systems. This is a practical concession: providers do not have to police licensees, but once they know of an issue, they need to act fast.

“Latent” and “manifest” definitions deleted. The prior statutory definitions of these terms have been deleted. SB 1000 focuses on provenance data in standardized, interoperable formats, aligning the law more closely with the C2PA ecosystem and similar technical standards.

Why This Matters 

If you create GenAI tools: The elimination of the one-million-user threshold means compliance with this law isn’t just a big tech problem anymore. Under SB1000, any publicly accessible GenAI system in California must embed latent provenance data and provide (or link to) a free verification tool.

If you license GenAI systems: SB1000 changes the landscape of license agreements to allow licensors to terminate access within 72 hours if licensees modify the system out of compliance. 

If you’re a platform: Large online platforms (2M+ monthly users) already face separate obligations under AB 853 (effective January 1, 2027) to detect and display provenance data. SB 1000’s expansion of who must embed that data means there will be significantly more provenance-tagged content flowing through your platform, which strengthens the case for investing in detection infrastructure now.

If you’re evaluating AI investments or M&A targets: Compliance with the California AI Transparency Act is now a concrete operational cost for any company shipping a public-facing GenAI product. Existing due diligence efforts could benefit from assessing whether targets have the technical capability to embed interoperable provenance metadata and whether their license agreements contemplate the 72-hour termination framework.

___________

If you have any questions about this Legal Briefing, please feel free to contact any of the attorneys listed or the Eversheds Sutherland attorney with whom you regularly work.