Do You Need to Label AI-Generated Content?

Yes, in specific situations, and the requirement depends on where your users are and what kind of content you are publishing. As of August 2026, no single global law forces every AI-generated image, video, or article to carry a label. But three separate rules now apply in parallel: the EU AI Act's Article 50 transparency obligations, California's AI Transparency Act, and FTC rules against undisclosed AI-generated reviews and endorsements. Each covers different content types, different company sizes, and different deadlines. Here is what is actually required, by whom, right now.

Two kinds of labeling get lumped together in most coverage of this topic, and separating them matters. Machine-readable marking is invisible: metadata or a watermark embedded in a file so a detector can identify it as AI-generated. Visible disclosure is a label a human reader actually sees, like a caption reading AI-generated. Some laws require one, some require both, and some only require you to offer the option to the user. Conflating the two is where a lot of writing on this subject goes vague.

The EU AI Act's Article 50 is the actual content-labeling rule

Article 50 is the specific provision in the EU AI Act that governs labeling of AI-generated content. It applies from 2 August 2026, according to the European Commission's own page on the obligation. It covers four things: telling people when they are interacting directly with an AI system, marking AI-generated or manipulated content in a machine-readable format, disclosing deepfakes in a clear and distinguishable way, and labeling AI-generated text published on matters of public interest that has not gone through human review.

A grace period applies only to the marking and detection obligation, and only for systems already on the market before 2 August 2026: those get until 2 December 2026 to comply. Everything else under Article 50 applies from the August date with no extension. For a deeper walkthrough of who counts as a provider versus a deployer under this article, see a closer look at the EU AI Act's transparency rules.

Why this gets confused with a different, later deadline

Article 50 is frequently mixed up with a separate part of the same law. The AI Act's Annex III rules, which classify certain systems as high-risk (used in employment, credit scoring, education, and similar contexts) and impose heavier compliance duties, were pushed back by a package known as the Digital Omnibus. That package entered into force on 27 July 2026, and it defers the Annex III high-risk rules to 2 December 2027 for standalone systems, and to 2 August 2028 for AI embedded in products already covered by EU product-safety law, according to the European Commission's own announcement that the Omnibus had entered into force. That same page lists no change to Article 50.

In other words, the content-labeling obligation and the high-risk classification deadline are two different parts of the same law, moving on two different timelines. Treating them as one deferred date, which a lot of 2026 coverage does, is the single most common inaccuracy in this space right now. For the mechanics of the high-risk deferral itself, see the AI Act's high-risk deadline explained separately.

California's AI Transparency Act (SB 942)

California has its own labeling law, the AI Transparency Act, known as SB 942. It applies to a covered provider, defined in the bill text as a business that creates a generative AI system with more than 1,000,000 monthly visitors or users, publicly accessible in California. The law was originally set to take effect 1 January 2026, but a later bill, AB 853, delayed the operative date to 2 August 2026, aligning it with the EU timeline.

It requires two things from a covered provider. First, mandatory latent disclosure: hidden metadata embedded in AI-generated image, video, or audio content that identifies the provider and system. Second, an option for manifest disclosure, meaning the provider must let users choose to add a visible label, though the label itself is not mandatory on every piece of content. Text-only output is not covered by the watermarking requirement at all. Covered providers must also offer a free public detection tool, and a separate hosting-platform obligation is scheduled to begin 1 January 2027.

Is there a federal US requirement to label AI content?

No comprehensive federal law requires AI-generated content to carry a label. What exists instead is enforcement under existing deceptive-practices authority. The FTC updated its Endorsement Guides in 2023 to explicitly cover AI-generated endorsements and synthetic reviewers, and finalized a rule in 2024 banning the creation, sale, or distribution of reviews and testimonials that misrepresent authorship, including reviews written by AI and presented as coming from a real person who does not exist.

That rule targets deception around reviews and endorsements specifically. It does not create a general AI-generated label requirement for ordinary content, articles, images, or marketing copy outside that context. If you advertise or sell into the EU as well, note that the two regimes are not equivalent: Article 50 is a content-labeling rule, while the FTC's authority here is a deception rule that happens to reach AI when it is used to fake authenticity.

What's required, by whom, right now

What this means if you build or publish with AI

A few practical points follow from the above. Check thresholds before assuming a law applies to you: SB 942 only binds covered providers with over a million monthly California users, which means most indie apps and small tools built on top of someone else's model are not the covered provider themselves, the underlying model vendor usually is. Check content type as well. Watermarking and labeling rules in both the EU and California focus heavily on image, audio, video, and deepfakes. Plain AI-assisted text is treated differently, and in the EU only text on public-interest matters without human review triggers a specific disclosure duty.

Absence of a legal mandate does not mean absence of risk. Advertisers still fall under the FTC's deception authority if undisclosed AI content misleads a buyer, and search visibility carries its own separate exposure, worth checking against the guidance on whether Google penalizes AI-generated content. Overstating what a tool actually does or how much of it is AI, sometimes called AI washing, creates a related but distinct reputational and legal problem, separate from any labeling statute.

For freelancers and agencies, the labeling question often shows up as a client relationship issue before it shows up as a legal one, which is a separate decision from what regulation requires, covered in whether you should tell clients you use AI. And because this sits inside a broader, still-shifting set of rules, it's worth tracking against the AI risks founders and builders are watching more generally.

FAQ

Do I have to label AI-generated content in the United States? There is no general federal law requiring it. The FTC can treat undisclosed AI content as deceptive in specific contexts, mainly reviews, testimonials, and endorsements, under its updated Endorsement Guides and its 2024 fake reviews rule. Outside advertising and reviews, no blanket US labeling mandate currently exists.

Does the EU AI Act require me to label AI-generated images or text? Yes, under Article 50, which applies from 2 August 2026. Generative AI content must be marked in a machine-readable format, deepfakes must be disclosed clearly, and AI-generated text on public-interest topics must be labeled unless it went through human editorial review. A grace period until 2 December 2026 applies only to the marking obligation, and only for systems already on the market.

What is the deadline for the AI Act's high-risk rules after the Digital Omnibus? The Digital Omnibus, in force since 27 July 2026, pushed the Annex III high-risk classification rules to 2 December 2027 for standalone systems and 2 August 2028 for systems embedded in already-regulated products. This is a separate deadline from Article 50, which was not deferred.

Does California's AI Transparency Act apply to small businesses or solo developers? Generally no. It applies only to covered providers, meaning generative AI systems with more than 1,000,000 monthly visitors or users publicly accessible in California. Most small businesses using someone else's AI tools are not the covered provider under the statute.

What happens if I don't label AI-generated content? It depends entirely on jurisdiction and context. In the EU, noncompliance with Article 50 can draw administrative fines. In the US, there is no general labeling penalty, but undisclosed AI content that misleads consumers can still trigger FTC enforcement or state consumer-protection claims depending on how it is used.