Your AI Content Now Has to Confess It's Fake

Last week we looked at the rule that your chatbot has to tell people it is AI. This week's rule is its close cousin, and it matters to anyone who creates content: marketers, agencies, media, and creators. If your work involves AI-generated images, video, audio, or certain kinds of text, you now have a duty to make that AI involvement visible to your audience. Let us break down exactly what has to be labelled, what does not, and how to do it.

What the rule actually says

Article 50 of the EU AI Act requires that AI-generated content be disclosed as such, so that people are not misled into thinking something artificial is real. The obligation focuses on two specific categories. The first is deepfakes, defined as AI-generated or manipulated images, audio, or video that resemble real people, objects, or events and could falsely appear authentic. If you publish one, you must clearly disclose that it is artificially generated or manipulated. The second is AI-generated text published to inform the public on matters of public interest, which must be disclosed as AI-generated unless a human took editorial responsibility for it.

The duty falls on the deployer, meaning the organisation that professionally publishes the content and puts it in front of people, whether or not you built the underlying AI. So a marketing team using a third-party tool to generate a realistic video of a person is responsible for labelling it, even though they did not build the model.

Who has to comply

This applies to any business or individual professionally publishing AI-generated content to an audience that includes people in the EU. If you run marketing campaigns with AI-generated visuals, produce media with synthetic voices or avatars, create AI video for clients, or publish AI-written articles on current affairs, you are in scope. As with every part of the AI Act, geography does not exempt you: a business based anywhere in the world is caught whenever its AI-generated content reaches people in the EU. This is the same extraterritorial reach, the Brussels Effect, that runs through the whole regulation.

What compliance actually looks like

The disclosure has to be clear and noticeable to the person seeing the content. For a deepfake video, that means a visible on-screen label or an opening disclaimer; for images, a visible mark; for audio, an audible or clearly stated disclosure. To make this easier, the European Commission has published three official, free-to-use icons, a basic AI mark, a "fully AI-generated" mark, and a "partially AI-modified" mark, available in standard image formats with no attribution required. Commission testing found that pairing an icon with a short text label improves recognition on every measure, so the recommended approach is icon plus a few words, not an icon alone.

There are sensible carve-outs worth knowing. Content that is evidently artistic, creative, satirical, or fictional attracts only a minimal, non-intrusive disclosure rather than a prominent one, so a label does not have to ruin the creative work. Clearly fantastical or physically impossible content, a dragon, a person flying unaided, falls outside the deepfake definition altogether. And AI-assisted text escapes the labelling duty where it went through genuine human editorial review with a named person taking responsibility for the publication.

What is at stake

The penalties mirror the rest of the transparency rules: non-compliance can draw fines of up to 15 million euros or 3 percent of total worldwide annual turnover, whichever is higher, with a lower figure applied proportionally to smaller businesses. Enforcement sits with national market surveillance authorities across the EU. Beyond the fine, there is a trust dimension: audiences increasingly expect transparency about AI, and being caught publishing unlabelled deepfakes carries a reputational cost that can outlast any penalty. Labelling is quickly becoming both a legal duty and a mark of credibility.

Your action this week

Here is the step to take. Make a list of every place your business publishes AI-generated content externally, marketing visuals, social media, video, audio, and any AI-written public-interest text. For each, ask two questions: is this a deepfake or realistic synthetic media, and if it is text, did a human take editorial responsibility? Where the answer means a label is required, add a clear disclosure now, ideally one of the official EU icons paired with a short text note. If you use third-party AI tools, remember the labelling duty sits with you, the publisher, not just the tool. This small step keeps you compliant and, increasingly, signals credibility to your audience. Next week, we look at how to classify your AI system into the right risk tier, the step that determines which of the heavier rules apply to you at all.