AI Act Article 50 deepfake labelling rules have applied since 2 August 2026. If you use an AI system professionally to create or manipulate images, audio or video that resemble real people, places or events and could falsely appear authentic, you must disclose that the content is artificially generated or manipulated. Providers of generative AI tools must also mark outputs in a machine-readable way, and people chatting with an AI must be told it is not a human. Breaches can be fined up to €15 million or 3% of worldwide turnover.

This guide explains the EU AI Act transparency obligations for adult platforms, creators, affiliates and tool builders. It is general information, not legal advice.

The four AI Act transparency obligations in Article 50

Article 50 of the AI Act (Regulation (EU) 2024/1689) sets out separate duties for "providers" (who develop or place an AI system on the market) and "deployers" (who use one under their authority for professional purposes).

DutyWhoWhat it means
AI interaction disclosure (50(1))ProviderDesign chatbots and companions so people are told they are interacting with AI, unless it is obvious
Machine-readable marking (50(2))Provider of generative AIMark synthetic audio, image, video and text outputs so they are detectable as AI-generated, for example with watermarks or metadata
Emotion recognition and biometric categorisation (50(3))DeployerInform people exposed to these systems
Deepfake and public-interest text disclosure (50(4))DeployerDisclose that deepfake content is artificially generated or manipulated; disclose AI-generated text published to inform the public on matters of public interest unless it has had human editorial review

The information must be clear and distinguishable and given at the latest at the time of first interaction or exposure, and must meet accessibility requirements.

What counts as a deepfake under the AI Act

The AI Act defines a deep fake 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. In practice:

  • Usually covered: a realistic AI video of a real performer, a face swap, a voice clone, a photorealistic image of an identifiable real person, or a realistic scene presented as real footage.
  • Grey area: photorealistic images of entirely fictional people. They may not "resemble existing persons", but they can still appear authentic. Many platforms label them anyway.
  • Usually outside the definition: obviously stylised art, cartoons and minor edits such as colour correction, though the line can move with realism.

Labelling is not a licence. Article 50 tells you how to disclose lawful synthetic content. It does not make non-consensual sexual deepfakes of real people lawful. Creating or sharing them can be a criminal offence in the UK and across the EU; see our guide to intimate image abuse and deepfake laws.

The artistic and satirical carve-out is narrow

Where deepfake content forms part of an evidently artistic, creative, satirical, fictional or analogous work, the obligation is reduced, not removed. You still have to disclose the existence of generated or manipulated content, but in an appropriate way that does not hamper the display or enjoyment of the work, for example a credit line or a discreet on-screen notice.

Two points matter for adult businesses:

  • "Evidently" means the creative nature must be obvious to the audience. Content marketed as real, or designed to look like genuine footage of a real person, will struggle to qualify.
  • Even when the carve-out applies, rights of the people depicted, including image rights, data protection and consent, still apply.

Who must label: platforms, creators, affiliates and tool builders

Creators and studios

If you use AI tools professionally to generate or alter realistic images, video or voice, you are a deployer. Label posts, previews and paid content that are AI-generated or substantially AI-manipulated. Our guide to AI content on creator platforms covers how platforms are handling this in their own terms.

AI chat and companions

Article 50(1) is a provider duty, but if you run an AI companion, an AI "chatter" replying to fans or a bot on your site, the system must make clear to users that they are talking to AI. Platform rules and consumer law already discourage presenting AI replies as the creator personally. See also our guide to AI companion affiliate programs.

Affiliates and advertisers

If you generate realistic AI images for banners, landing pages or reviews, label them. Advertising codes add their own rules on misleading imagery; read our guide to AI images and filters in ads.

Generative tool providers

If you build or white-label an image, video or voice generator, the 50(2) marking duty is yours. Under the transitional rule added by the Digital Omnibus on AI (Regulation (EU) 2026/1744), generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to implement machine-readable marking; new systems should comply from launch.

How to label AI images and video in practice

  1. Visible label: a clear, consistent tag such as "AI-generated" on the content and in previews, placed where users see it before or at first view.
  2. Metadata: keep provenance metadata from your tools intact and, where possible, add content credentials. Do not strip it during compression or upload.
  3. Upload flow: on platforms, add a mandatory "contains AI-generated or manipulated content" question for uploaders, and apply labels automatically.
  4. Detection: use the marks applied by generative tools to catch unlabelled synthetic uploads.
  5. Policy: state in your terms how AI content must be labelled and what happens if it is not, then enforce it through your content moderation policy.

In June 2026 the Commission published a voluntary Code of Practice on marking and labelling AI-generated content to help providers and deployers show compliance. Following it is not mandatory, but it is the clearest benchmark available.

Enforcement and penalties

Article 50 is enforced mainly by national market surveillance authorities designated by each Member State, with the Commission's AI Office co-ordinating and supervising general-purpose AI models. Breaches of the transparency obligations can attract fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher, with lower caps for SMEs and start-ups.

Worked example (hypothetical): a studio publishes a realistic AI-generated trailer featuring a digital double of a contracted performer. The performer consented and the contract covers AI use, so the content is lawful, but the trailer still needs a clear "AI-generated" disclosure when shown to EU viewers. Keeping the consent record and the label together is what protects the studio if a regulator, payment partner or platform asks.

Synthetic media rules outside the EU, including the UK

The AI Act applies to deployers established in the EU and to providers and deployers elsewhere when the output is used in the EU, so a UK or US site with EU users should assume it is in scope. The UK has no equivalent labelling law, but creating sexually explicit deepfakes of adults without consent became an offence in England and Wales on 6 February 2026, and the Crime and Policing Act 2026 targets nudification tools. The UK advertising regulator also expects ads not to mislead through AI imagery.

Frequently asked questions

Do I need to label AI-generated images of fictional people?

If they are photorealistic and could appear authentic, labelling is the safer course and many platforms require it. Clearly stylised or illustrated content is lower risk.

Is a hashtag enough?

A clear hashtag or caption can be part of it, but the disclosure must be distinguishable and visible at first exposure. A tag buried among dozens of others is unlikely to be enough.

Who is fined if an uploader fails to label?

The uploader is normally the deployer. A platform's own exposure is more likely to come from the Digital Services Act and its terms of service, so enforce labelling rules consistently.

Next steps

Audit where you use generative AI, add labels and preserve metadata, and update your upload rules. If you license synthetic or AI-assisted content, check providers in our content providers directory for clear labelling and consent terms.