California AI Transparency Act (SB 942)

The California AI Transparency Act (SB 942) is a California content-provenance law requiring large generative AI providers to offer a free public AI-detection tool, embed machine-readable provenance disclosures in generated image, video, and audio content, and offer users a visible disclosure option. It was signed by Governor Gavin Newsom on September 19, 2024, and took effect January 1, 2026. Secondary analyses describe it as the first US statute to mandate machine-readable provenance disclosure in generative AI outputs.

Status and timeline

SB 942 was introduced in the 2023–2024 Regular Session and authored by Sen. Josh Becker (D-Menlo Park). Newsom signed it on September 19, 2024, with an effective date of January 1, 2026. Amendments enacted through AB 853, signed in October 2025, narrowed the statute's scope and clarified its contract requirements (Source: bill text https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB942).

Scope and definitions

The act applies to "Covered Providers" — persons that create, code, or otherwise produce a generative AI system that has more than 1,000,000 monthly visitors or users and is publicly accessible in California. The 1,000,000-user threshold captures the largest consumer-facing generative AI services (ChatGPT, Gemini, Claude, Copilot, Meta AI, and image or video tools such as DALL-E, Midjourney, Sora, Veo, Runway, and ElevenLabs) while excluding smaller, enterprise-only, or API-only providers without a consumer frontend.

Key provisions

Free AI-content detection tool. A Covered Provider must make publicly available, at no cost, a tool that allows any person to upload image, video, or audio content (or combinations) and determine whether the content was created or altered by the provider's generative AI system. The tool must be accessible, support common file formats, disclose results in plain language, and include an explanation of how the detection works and its limitations.

Latent (machine-readable) disclosure. Covered Providers must automatically embed a "latent disclosure" in AI-generated or AI-altered image, video, or audio content. The disclosure must convey — directly or through a permanent internet link — the name of the Covered Provider, the name and version of the generative AI system, the time and date of generation, and a unique identifier. It must be implemented consistent with industry standards, implicitly aligned with C2PA / Content Credentials, and must be difficult to remove by ordinary means.

Manifest (user-facing) disclosure. Covered Providers must offer users the option to include a manifest, clearly visible, disclosure identifying the content as AI-generated. The manifest disclosure must be clear, conspicuous, appropriate for the medium, and difficult to remove.

Third-party licensing and contract requirements. If a Covered Provider licenses its generative AI system to another entity that removes the Covered Provider's required disclosures, the Covered Provider must terminate the license within 96 hours of learning of the removal. Secondary analyses describe this as the first US statute to impose cascading contractual obligations on downstream deployers of generative AI.

Enforcement and penalties

The act sets a civil penalty of $5,000 per violation per day. Enforcement rests with the California Attorney General, a city attorney, or county counsel. There is no private right of action.

Comparison with other legislation

SB 942 mandates both machine-readable (latent) and human-visible (manifest) disclosure for generative AI providers serving more than 1,000,000 users, pairing required watermarking with a free public detection tool. Secondary analyses describe it as the first US statute to mandate machine-readable provenance disclosure in generative AI outputs, converging on the C2PA standard and operationally compatible with EU AI Act Article 50.

Relative to EU AI Act (Regulation 2024/1689) Article 50, SB 942 reaches a narrower set of providers (consumer-scale only) but is more prescriptive on the detection-tool side.

Implementation and contested questions

Several aspects of the statute's operation have drawn commentary as contested or unsettled.

On the technical feasibility of robust watermarking, the academic literature (including the 2023 arXiv work On the Reliability of Watermarks) holds that watermarks can be removed or forged; SB 942's "difficult to remove" standard rests on that ground. Research from Scale and MIT documents weaker watermark robustness in audio than in image content, and practitioners expect early SB 942 enforcement to focus on image and video.

On detection-tool liability, a tool that returns a false negative for harmful content such as non-consensual intimate imagery (NCII) exposes the provider to reputational and potentially Section-5-adjacent risk, creating incentives toward conservative detection.

On the 96-hour license-termination duty, large model APIs have thousands of downstream users, making real-time monitoring of disclosure removal across the stack operationally difficult. The duty also overlaps with the federal NCII takedown obligation under the TAKE IT DOWN Act — Source Summary, whose 48-hour platform takedown timeline runs alongside SB 942's 96-hour license-termination timeline; platform-provider coordination across the two is not well specified.

On open-weights models, providers that release open weights (Meta Llama, Qwen, DeepSeek) cannot enforce downstream watermarking, a gap the statute does not address.

On federal preemption, Executive Order 14365 — Ensuring a National Policy Framework for AI and the America's AI Action Plan posture against state AI transparency mandates; the resulting enforcement posture remains contested. Secondary analyses describe SB 942 as a template likely to be adopted by other states and referenced in EU AI Act Article 50 implementing acts, and as part of a shift toward machine-readable provenance as a governance tool alongside human-visible disclosures.

Relationships

• related: EU AI Act (Regulation 2024/1689) — Art. 50 parallel on deepfake/GenAI marking
• related: TAKE IT DOWN Act — Source Summary — adjacent federal NCII/deepfake regime
• related: State Deepfake Statutes (MN, WA, TX, CA) — political-deepfake state laws use manifest disclosure
• related: California AB 3030 (Healthcare AI Disclosure), California SB 896 (Generative AI Accountability Act), California SB 53 — Transparency in Frontier AI Act — CA GenAI package
• related: US AI Regulatory Approaches Compared — adds the "provenance / watermarking" dimension

Sources

• Bill text: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB942
• Secondary: Jones Day (Oct 2024); Troutman Pepper Locke (Oct 2025 on amendments); Orrick (Jan 2025 contract guide); Norton Rose Fulbright Data Protection Report (Sep 2024); Cooley (Oct 2024); CalMatters Digital Democracy