EU AI Act Deepfake Labeling: What Creators Must Do

EU AI Act deepfake labeling means that, since August 2, 2026, anyone who publishes AI-generated or manipulated image, audio or video content that could falsely pass as real, as part of a professional activity, must disclose that the content is artificial. For creators, that usually means a visible label on the post or in the video itself, shown no later than the moment a viewer first sees it, on top of whatever invisible watermark the AI tool adds.

Sources, checked October 2026: Article 50 on the European Commission's AI Act Service Desk (consolidated text as of July 2026), the Commission's FAQ on Article 50 transparency obligations (last updated July 24, 2026), the Code of Practice on Transparency of AI-generated Content and the full regulation on EUR-Lex. This is general information, not legal advice; for a specific campaign, ask a lawyer in the member state where you operate.

This page is the creator-side summary: what you have to do, what the AI tool company has to do, which content is caught, and a checklist you can run before you post. It sits under our wider AI influencer legal and ethics guide, which covers advertising disclosure, copyright and likeness rights across countries. If you only want the platform side, our guide to Instagram AI labels explains how Meta applies its own tags.

Who this applies to: are you a deployer?

The AI Act splits responsibility between two roles. A provider builds an AI system and puts it on the market: the image generator, the video model, the voice cloning service. A deployer is the person or business that uses an AI system under its own authority. The deepfake labeling duty in Article 50(4) falls on deployers.

The key carve-out is personal use. The Commission's FAQ says that when a person uses an AI system in a personal capacity, for example to make a deepfake and share it on social media, that is a personal, non-professional activity and falls outside the AI Act. But if it is an activity through which the person gains an economic benefit on a regular basis, or it is part of a business, trade, occupation or freelance work, the person is a deployer.

Most readers of this site are on the deployer side of that line. If you run an AI influencer account that earns from brand deals, subscriptions or affiliate links, make content for clients, or sell AI images, you are acting professionally. When a business deploys the system, its employees and contractors working under its control are not separate deployers; the business is. A freelancer working for themselves is their own deployer.

Location matters less than you might expect. Article 2(1)(c) brings in providers and deployers established outside the EU when the output of the AI system is used in the EU. A creator based in the US or UK whose deepfake content is aimed at, and shown to, audiences in the EU should assume the rule reaches them.

How EU AI Act deepfake labeling works end to end

Two layers of transparency stack on top of each other. The provider marks the file so machines can detect it. The deployer tells the human viewer. Neither replaces the other.

The Commission is explicit on one point that trips up creators: deployers cannot rely on the machine-readable marking embedded by the provider to meet their disclosure duty. A watermark or metadata tag that only software can read does not tell a person anything. The label has to be understandable and perceivable by people without special tools or extra clicks.

What counts as a deepfake, in plain terms

Article 3(60) defines a deepfake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. The Commission's guidance breaks that into three tests, and all three must be met:

• Resemblance. A high level of similarity between the content and the thing it simulates.
• Existing or plausible. The simulated person, object, place or event exists, could plausibly exist, or could plausibly have existed. This is the part that catches AI influencers: a realistic person who never existed is still a plausible person.
• False appearance of authenticity. The content could mislead someone about whether it is real. Context counts here: the level of resemblance, the message, where the content is shown and what that audience expects.

The Commission gives a useful counterexample: background scenes, special effects and routine pre- and post-production in standard film-making are unlikely to make content falsely appear authentic. The same logic helps creators. A stylised illustration, an obviously fantastical scene, or a cartoon mascot is unlikely to be a deepfake. A photorealistic selfie of an AI persona at a real café, a cloned voice of a real person, or an edited clip that changes what someone said is very likely to be one.

"Likely" is doing real work in that comparison. The test is contextual, and a regulator will judge your specific post. If the honest answer to "could a follower think this is a real photo or recording?" is yes, label it. Labelling a borderline post costs you nothing; missing one costs you trust and, potentially, a fine.

What does EU AI Act content labeling require?

Article 50 has four separate transparency duties, and only some of them are yours. The table maps common creator situations to the role that carries the duty.

For deployers, the rule on how to disclose sits in Article 50(5): the information must be given to the people concerned in a clear and distinguishable manner, at the latest at the time of first exposure, and must meet applicable accessibility requirements. In practice that rules out a disclosure buried at the end of a long caption, hidden in a profile bio, or placed only in alt text.

Text has its own rule. If you publish AI-generated or manipulated text with the purpose of informing the public on matters of public interest, such as politics, public health, consumer safety or economic developments, you must disclose it unless the text went through real human review or editorial control and a person or company holds editorial responsibility. Spell-checking does not count as review. Most creator captions about your day, a product or a tutorial are not public-interest information in this sense, but a newsletter or explainer on a current policy issue might be.

The artistic, satirical and fictional exception

Article 50(4) softens, rather than removes, the duty for content that forms part of an evidently artistic, creative, satirical, fictional or analogous work. In that case you only need to disclose the existence of the generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work.

Two words matter: "evidently" and "appropriate". A short film that is clearly fiction can carry a credit-style disclosure rather than a permanent banner. A parody sketch can say it is AI-made in the opening card or the caption. But an AI influencer posting everyday lifestyle content is not evidently fictional to a casual viewer, which is the whole point of a realistic persona, so the lighter treatment rarely applies to it.

Key dates and enforcement

The Act was amended in 2026 by the Digital Omnibus on AI, and the Commission's Service Desk marks the changed passages in its consolidated text. The deployer labeling duty in Article 50(4) still applies from August 2, 2026; the December 2, 2026 grace period the Commission describes covers only the provider-side marking duty for systems already on the market. Content you generated before August 2, 2026 does not have to be labelled retroactively, although the Commission encourages deployers to do it where possible.

Enforcement sits mainly with national market surveillance authorities in each member state. The AI Office has a narrower role, for example where an AI system is built on a general-purpose model by the same company or is integrated into a very large online platform. The Commission says fines for breaching these rules can reach 15 million euros or 3 percent of total worldwide turnover for the preceding financial year, with proportionality for small and medium-sized businesses.

What does AI Act transparency mean for creators in practice?

From running AI persona accounts, the practical answer is a labelling habit that survives reposts, crops and different platforms. The steps below are the routine we recommend: it meets the visible-disclosure idea in Article 50 and also lines up with what the major platforms ask for.

The on-content label matters most because it travels. A caption disappears the moment someone screenshots or reposts your image, and platform labels do not always carry across apps. Article 50(5) asks for disclosure at first exposure, and for many viewers first exposure is a repost.

Wording does not need to be legalistic. "AI-generated image", "This video uses an AI-generated voice" or "Virtual creator: images made with AI" are all clear. Avoid vague terms like "digital art" or "enhanced" for photorealistic content, since they do not tell a viewer that the person or scene is not real.

Does the EU AI Act apply to social media content?

Yes, when the poster is acting professionally. The Commission's own example of a personal, out-of-scope use is a person generating deepfakes and sharing them on social media in a private capacity. Flip that to a monetised creator account, an agency running client pages, or a brand account, and the same post becomes deployer activity that must be labelled.

The AI Act also does not replace platform rules. Meta, TikTok and YouTube each have their own AI disclosure settings and their own penalties for unlabelled realistic content, and those apply whether or not you are in the EU. Article 50(6) adds that the AI Act is without prejudice to other transparency duties in EU or national law, such as consumer protection rules on advertising. A sponsored post that uses an AI persona can need two disclosures at once: one that it is an ad, and one that the content is AI-generated.

Platform labels are worth switching on, but treat them as a second layer. Labels are applied inconsistently and can be removed or never triggered, and the Commission is clear that the visible disclosure is the deployer's responsibility. For how Meta's own tag behaves, see our Instagram AI label guide.

The Code of Practice: should creators care?

The Code of Practice on Transparency of AI-generated Content was drawn up by independent experts in a process run by the AI Office. Its first section covers providers and marking; its second covers deployers and the labelling of deepfakes and AI text. The Commission says the code has been assessed as adequate by both the Commission and the AI Board, so signatories can rely on it to show compliance.

Signing is voluntary, and an individual creator is unlikely to sign. It is still worth reading the deployer section, because it is the most concrete picture of what "clear and distinguishable" labelling looks like to regulators. Those who do not sign must show compliance by other adequate means and, as the Commission notes, may face more requests for information.

Common mistakes we see

• Disclosing only in the bio. A profile line saying "AI creator" helps, but people often meet a post in a feed, a search result or a repost, never visiting the profile.
• Hiding the label after the cut. A disclosure at the bottom of a long caption is not clear at first exposure for anyone who does not tap "more".
• Treating realism as fiction. A persona's backstory is fiction to you; to a follower scrolling past a realistic selfie it looks like a real person.
• Forgetting voice. Audio deepfakes are in scope. A cloned voice-over needs an audible or on-screen disclosure, not just a text note.
• Labelling once, then reusing. Clips cut from a labelled video for Stories or Shorts often lose the label. Re-check every derivative.
• Mixing up ad and AI disclosure. "#ad" says you were paid. It says nothing about whether the content is real.

If you are building or scaling an AI persona and want the whole workflow, including character design, consistent faces, disclosure and brand deals, our AI Influencers program covers it step by step, with the labelling routine built into the posting process rather than added at the end.

How this fits with other deepfake and likeness rules

Article 50 is a transparency rule. It does not give anyone a right to stop you using their face, and it does not make a labelled deepfake lawful if it breaks other laws. Using a real person's likeness still raises consent, privacy and personality-rights issues under national law, which our guide to AI likeness rights covers. Face-swapping a real person is the riskiest case; our AI face swap guide explains where the lines are. And if you run AI creators in paid campaigns, our AI UGC ads guide covers the advertising side.

Your Article 50 checklist

None of this needs special software. A text layer in your editor, a caption template and a habit of checking each clip before it goes out cover nearly every creator case. Review the Commission's FAQ every few months: the guidelines and the code are new, and both the wording and national enforcement practice will develop.

EU AI Act deepfake labeling: FAQ