California's AI Transparency Act Takes Effect Today: What SB 942 Means for Every Major AI Company

Today, August 2, 2026, a new compliance reality hits Silicon Valley. California’s SB 942—the AI Transparency Act—became operative this morning, imposing binding legal requirements on every major generative AI company that reaches California users. For OpenAI, Google, Anthropic, Meta, xAI, and any other provider with more than one million California monthly active users, the grace period is over.

What the Law Requires

SB 942 creates three interrelated obligations for covered AI providers:

Latent provenance watermarks. Providers must embed C2PA-compatible machine-readable watermarks in all AI-generated images, video, and audio. These hidden signatures encode metadata about the generating model, timestamp, and provider identity. Unlike visible labels that users can screenshot and strip, latent watermarks are designed to survive format conversion and social sharing.

Free public detection tools. Providers must offer publicly accessible tools that allow anyone—journalists, researchers, ordinary users—to check whether a piece of media was AI-generated and identify its origin. The detection tool must be free to use, with no account creation required.

User-facing disclosure options. Providers must give users the ability to add visible “AI-generated” labels to content they create, if they choose. This is an opt-in for users, not a mandatory label on all outputs.

Equally significant is what the law prohibits: stripping or removing embedded provenance data is now illegal in California. Building or distributing tools designed to erase AI watermarks is also prohibited—a provision aimed directly at the growing cottage industry of AI-detection evasion tools.

A Two-Year Journey to This Moment

SB 942 was signed by Governor Gavin Newsom on September 19, 2024—the culmination of a legislative push that had been building since the first wave of convincing deepfakes disrupted the 2024 election cycle.

The original operative date was January 1, 2026, but AB 853, signed in October 2025, pushed enforcement back to August 2, 2026. That date was not chosen arbitrarily: it aligns precisely with the EU AI Act’s enforcement deadline for transparency requirements on high-risk AI systems. The alignment was deliberate, intended to force a global compliance moment rather than allowing companies to maintain different content labeling practices for California versus European users.

AB 853 also significantly expanded the law’s scope beyond the original SB 942 text. The amendment added three new categories of covered entities: large online platforms with more than two million unique monthly California users (catching platforms that distribute AI-generated content even if they don’t create it); generative AI hosting platforms (catching cloud providers that host third-party AI models); and capture device manufacturers (requiring smartphone makers to implement provenance support at the hardware level).

Who Is Covered—and What They Must Do

The law’s coverage is deliberately broad. The one-million-user threshold applies to California monthly active users, not global users, and given California’s 40 million residents and high rate of AI tool adoption, virtually every major AI product crosses this threshold.

This means the requirements apply across the full stack: OpenAI must watermark ChatGPT-generated images; Google must watermark Imagen and Gemini outputs; Meta must watermark content from its AI image generator inside Instagram and WhatsApp; Apple’s AI-generated image features fall under the capture device manufacturer provision; and major cloud platforms hosting third-party AI APIs must ensure their infrastructure supports C2PA metadata passthrough.

The hospitality for startups is lower than it looks. A startup with a breakout AI image product can hit one million California monthly users quickly—at that point, SB 942 compliance becomes a legal requirement within the same growth cycle as the initial success.

The Enforcement Teeth

Penalties are $5,000 per day per violation, enforced by the California Attorney General’s office. “Per violation” means per piece of non-compliant content, not per company per day—a calculation that makes exposure extremely large for high-volume generators.

The AG’s office has not yet announced specific enforcement actions, but the California Department of Justice has a track record of selective, high-profile enforcement in technology cases. Early enforcement is likely to target egregious violations—stripped watermarks, missing detection tools—rather than edge cases in implementation.

Private right of action is notably absent from SB 942. Only the AG can bring enforcement actions, which limits the litigation risk that concerned some companies during the bill’s passage but also concentrates enforcement in a single office that must prioritize its cases.

The C2PA Question

SB 942’s requirement for “C2PA-compatible” watermarks is both its strength and its potential weakness. The Coalition for Content Provenance and Authenticity (C2PA) standard, now on version 2.1, has broad industry backing: Adobe, Microsoft, Google, Sony, and the BBC are all members. The standard is technically mature and has been deployed in production by Adobe (Content Credentials) and in limited capacity by OpenAI (DALL-E images).

However, C2PA adoption at the infrastructure level—meaning robust watermarks that survive social platform compression, screenshot extraction, and cross-format sharing—remains incomplete. Platforms like Twitter/X, TikTok, and many messaging apps do not currently preserve C2PA metadata. A user who downloads a watermarked image and reuploads it may inadvertently strip the provenance chain.

SB 942’s prohibition on actively stripping watermarks addresses the intentional evasion case, but the passive data loss from non-compliant platforms is a harder problem. Legal experts expect California to address this through follow-on regulations or AG guidance on what constitutes reasonable implementation.

The EU Alignment and Global Impact

California’s deliberate synchronization with EU AI Act Article 50—which requires disclosure that AI systems generated or manipulated content—creates the global compliance moment the law’s authors sought.

Companies must now maintain content provenance systems that satisfy both California and EU requirements simultaneously. Since these requirements are broadly aligned (both require metadata disclosure and user-facing labels), most companies will build a single global implementation rather than separate California and EU versions. This means SB 942 functionally sets content provenance standards for global AI deployment, not just California users.

The timing also coincides with the Trump administration’s notably more permissive approach to federal AI regulation. The White House yesterday missed its own August 1 deadline for frontier AI model disclosure requirements under Executive Order 14409, signaling low federal enforcement pressure. California is filling that vacuum—as it has on privacy, automotive safety standards, and environmental regulation—by setting de facto national standards through state law.

What to Watch

The law’s first year will be defined by three questions: How aggressively will the AG enforce? Will platforms update their infrastructure to preserve C2PA metadata? And will the hardware-level requirements for capture devices actually move Apple, Samsung, and Google to bake provenance into camera hardware at scale?

For developers and product teams building AI applications with media generation features, the message is clear: content provenance is no longer optional if you want California users. The clock started today.