AI Content Labeling Laws in 2026: What Must Be Disclosed Now

Publishing AI-generated content without disclosure is increasingly not just an ethical question—it's a legal one. By mid-2026, content creators, platforms, publishers, and marketers face a patchwork of AI labeling requirements across jurisdictions that have teeth. Fines have been issued. Investigations are ongoing. The "we didn't know the rules" defense is no longer credible.

Why This Became a Legal Issue

The regulatory push for AI content labeling came from two converging concerns. First, AI-generated deepfakes and synthetic media were influencing elections, creating reputational harm, and powering fraud at scale. Second, AI-generated journalism, marketing, and educational content was being published without disclosure, raising issues of trust, accuracy, and informed consent.

The legislative response has not been uniform—different jurisdictions have taken different approaches—but the direction is consistent: AI-generated or AI-manipulated content that has the potential to deceive must be labeled.

What the EU Requires

The EU AI Act includes content transparency provisions that are now fully in force. Under Article 50:

Synthetic media must be labeled. Content generated by AI that depicts non-existent persons, events, or places in a realistic manner must be marked as AI-generated. This applies to images, audio, and video.

Deepfakes must be disclosed. When existing real people are depicted in AI-generated or AI-manipulated audio or visual content, disclosure is mandatory unless it is "clearly art or satire."

Emotion recognition and biometric systems must disclose. Operators of AI systems that analyze people's emotions or categorize them biometrically must inform those people that such a system is in operation.

Chatbots must identify as AI. Users interacting with AI chatbot systems must be clearly informed they are interacting with AI, except in context where this is obvious.

The penalties for non-compliance under the AI Act are significant—up to €15 million or 3% of global annual revenue for transparency violations.

The United States: A Patchwork Landscape

The U.S. has not passed comprehensive federal AI content labeling legislation as of mid-2026. Several bills have been introduced in Congress, but none have reached the floor for a vote. What exists is a state-level patchwork and sector-specific rules.

California: SB 942 (California AI Transparency Act, effective 2024) requires large AI providers to make AI detection tools publicly available. AB 2655 requires large online platforms to label AI-generated election-related content in the 120 days before an election. Several additional bills extending these requirements are in the California legislative pipeline.

Texas and Florida have AI deepfake disclosure laws specifically addressing sexually explicit content and political advertising.

FTC guidance: The Federal Trade Commission has issued guidance treating undisclosed AI-generated content in commercial contexts as a potential deceptive trade practice under Section 5. The FTC's stance is that material connections to AI generation—particularly where it affects the perceived authenticity of reviews, endorsements, or testimonials—must be disclosed.

The FCC has moved to require disclosure of AI-generated voices in political advertising on broadcast media, following concerns about synthetic candidate audio in 2024 elections.

SEC guidance has addressed AI-generated financial disclosures and investment content, requiring disclosure of material AI use in research reports and investment communications.

China's Approach

China's Deep Synthesis Provisions (issued in 2022 and expanded since) require labeling of deepfakes and synthetic audio-visual content. Under these rules:

• AI-generated video and audio must carry visible watermarks or explicit text disclosure
• Platforms hosting such content must verify user identities and maintain records
• "Realistic-seeming" AI text must be labeled when published on platforms

China has been more aggressive about enforcement than Western jurisdictions in some respects, though the rules apply primarily within the Chinese internet ecosystem.

What Content Creators Must Do

The practical implications differ by content type:

Video and audio creators:

• AI-generated synthetic voices narrating content require disclosure in most jurisdictions with laws in force
• AI-generated video of real people—whether voices or likenesses—requires disclosure and in some cases consent
• AI-generated video of non-existent people should be labeled in the EU and is recommended practice globally

Writers and publishers:

• No jurisdiction currently requires disclosure of AI-generated text in general publishing (outside specific electoral and commercial contexts)
• However, the EU AI Act's transparency requirements for AI systems used to generate content that influences public opinion are being interpreted broadly by some national authorities
• Platform policies (Google, Meta, LinkedIn) now require labeling of AI-generated content in various contexts

Advertisers and marketers:

• FTC guidance makes undisclosed AI-generated testimonials and reviews a compliance risk in the U.S.
• The EU's Unfair Commercial Practices Directive has been interpreted to require disclosure of AI generation in commercial communications
• Influencer marketing using AI-generated likenesses or voice clones requires explicit disclosure

Political advertising:

• This is the most heavily regulated category globally. Multiple states in the U.S., the EU, and several other jurisdictions now require explicit labeling of AI-generated or AI-manipulated content in political advertising.

Platforms and Their Own Rules

Beyond government regulation, major platforms have their own disclosure requirements that creators must follow:

Google's search quality guidelines now treat undisclosed AI-generated content as a spam signal in certain contexts and require disclosure of AI use in content submitted through automated means.

YouTube requires creators to disclose when videos include AI-generated synthetic or "realistic altered or synthetic content," including digitally altered real people, non-existent people depicted realistically, or depictions of real events that didn't happen.

Meta's platforms require disclosure for AI-generated images and videos that could be mistaken for real photographs or footage.

The AI Transparency in 2026: What Companies Must Now Disclose article covers corporate disclosure requirements in depth.

Technical Standards for Disclosure

One challenge in AI content labeling is the lack of standardized disclosure formats. Several initiatives are working to address this:

C2PA (Coalition for Content Provenance and Authenticity) has developed technical standards for embedding content provenance metadata in files—allowing tools to detect and verify the origin of images, audio, and video. Major technology companies including Adobe, Microsoft, Google, and Apple have committed to implementing C2PA standards in their tools and platforms.

Watermarking research has advanced significantly. Google DeepMind's SynthID watermarking approach for text, images, and audio is being licensed to other companies. The challenge is that watermarks can be removed or degraded, so technical watermarking complements rather than replaces explicit disclosure.

What's Coming

The regulatory direction is toward more stringent requirements, not fewer:

• Federal U.S. legislation is expected to pass in some form in 2026 or 2027, likely addressing synthetic media and AI in political advertising specifically
• The EU is developing implementing regulations under the AI Act that will provide more specific technical requirements for synthetic media labeling
• Industry self-regulation through the C2PA and similar initiatives is accelerating—driven partly by regulatory pressure and partly by platform policy

For content creators and publishers operating globally, the safest approach in mid-2026 is to disclose AI generation in any context where a reasonable viewer might be deceived or where the content touches on sensitive domains including politics, finance, and health. This is good practice ahead of law in most U.S. contexts and required practice in the EU.