DEV Community

Mirage Cloud IA
Mirage Cloud IA

Posted on

Do You Have to Label AI Images and Text? The AI Act Explained

Almost every marketing team now uses an AI tool for images or copy, and since the summer there has been a labelling rule in force in the EU. The headlines are confusing, because one date, 2 December 2026, keeps appearing, and it is easy to conclude that nothing is required until then. For most small businesses the position is both narrower and more specific.

Who has which duty
Article 50 of the AI Act separates two roles. Providers are the companies that build and sell generative AI tools. Deployers are the businesses that use those tools, which is what most readers are.

➟ Providers must make sure that their systems' outputs, whether audio, images, video or text, are marked in a machine-readable way so that they can be detected as artificially generated. For systems already on the market before 2 August 2026, the Digital Omnibus gives them until 2 December 2026. That is the date in the headlines, and it is mainly a duty for the tool makers.
➟ Deployers must disclose deepfakes, and must label AI-generated or manipulated text that is published to inform the public on matters of public interest. Their duties have applied since 2 August 2026, with no extra grace period.

What counts as a deepfake
In the Act, a deepfake is AI-generated or manipulated image, audio or video content that resembles real people, objects, places or events and could falsely appear authentic. A realistic fake photo of an identifiable person, a cloned voice, or a scene that looks like footage of a real event are the clear cases. Obviously illustrated or stylised artwork, or a product mock-up that nobody would take for a photograph of a real event, is a different matter, and where content is evidently artistic or creative the disclosure can be kept light. Text is not a deepfake under the Act.

AI text that needs a label
The text rule is narrow. It applies to AI-generated or manipulated text published to inform the public on matters of public interest, such as news-style articles on politics, health or finance. It does not apply when a human has reviewed the text and someone takes editorial responsibility for it. Ordinary marketing copy, product descriptions and social posts are generally outside it, although the rules on misleading advertising still apply to what they say. The date that counts is the publication date, so text generated earlier but published after 2 August 2026 is caught.

How to label
The Commission published the final version of its Code of Practice on transparency of AI-generated content on 10 June 2026, and it and the AI Board have confirmed that following the Code is an adequate way to show compliance. Together with the Commission's guidelines, it points to a clear visual label shown at the first exposure to the content, using a common EU AI label or an equivalent. For images and video that means a visible tag or caption, and for audio a spoken or written disclosure at the start. Plain wording is enough, for example: Image generated with AI.

What to do this month

  1. List where you use AI to make images, video, audio or text that you publish.
  2. Label deepfakes and realistic synthetic people or events with a visible notice.
  3. Do not use AI to create realistic photos of customers, staff or events that did not happen. Beyond the AI Act, that is the misleading practices route.
  4. Ask your tool vendors how they mark outputs and whether the marking will be in place by 2 December 2026, and keep their written reply.
  5. Keep original files with their metadata and do not strip any marking.
  6. For public-interest articles, keep a record of the human review and of who is responsible for the content.
  7. Add one line to your content policy that says who may publish AI-generated material and how it is labelled.

Other rules that still apply
➟ Fake testimonials and reviews, AI-generated or not, are banned practices under consumer law.
➟ Copyright. French and EU copyright protects original works that reflect a human author's creative choices. Output generated with little or no human creative input is unlikely to be protected, so do not assume you can stop others from copying it. The law is still developing, so check the tool's terms on commercial use.
➟ Image rights and data protection. Do not generate or alter images of real people without permission. A face is personal data.
➟ Platform rules. Social networks have their own AI labels and policies. Follow them in addition to the law.
➟ Advertising. An AI-generated product image that shows something the product cannot do is misleading, whoever made the image.

Where AI helps
Once you have a rule, the labelling job is small. Mirage Cloud's changelog lists an image generation studio and a media library, so the question of how outputs are marked applies to it as it does to any other tool. Whatever you use, ask the vendor, label what you publish where the law requires it, and keep a human responsible for what goes out.

Top comments (0)