Blog · EU AI Act · Media
Label it or risk it: Article 50 and the new rules for AI-generated media
3 August 2026 · 5 min read · By LeadAI
When the EU postponed the high-risk parts of the AI Act, the transparency rules did not move. Article 50 has applied since 2 August 2026, and it reaches far beyond technology companies. If your organisation publishes AI-generated or AI-manipulated images, audio, video or certain kinds of text, this one is for you. See also our note on what the delay did and did not change.
What Article 50 asks for
Article 50 contains four duties. Two of them matter most for media and marketing.
| Duty | Who carries it | In plain English |
|---|---|---|
| Article 50(1) | Providers of interactive AI | Tell people when they are talking to an AI system, such as a chatbot. |
| Article 50(2) | Providers of generative AI | Mark outputs in a machine-readable way so they can be detected as AI-generated. |
| Article 50(3) | Deployers of emotion recognition or biometric categorisation | Inform the people exposed to these systems. |
| Article 50(4) | Deployers of generative AI | Label deepfakes, and AI-generated text published to inform the public on matters of public interest. |
A deployer is any organisation using an AI system under its own authority in a professional setting. That includes newsrooms, publishers, broadcasters, agencies, marketing teams and HR departments. If you use a third-party tool to make content you publish, the visible label is usually your job, even though marking the file is the tool provider's job.
The dates
- Article 50 applies from 2 August 2026.
- Generative AI systems already on the market before that date have until 2 December 2026 to meet the machine-readable marking duty in Article 50(2). This grace period applies to providers, not to your labelling duty as a deployer.
- The Commission published a final Code of Practice on marking and labelling AI-generated content on 10 June 2026. Following it is voluntary, but it shows what regulators are likely to expect, and the legal duties in Article 50 apply whether or not anyone signs it.
What the Code of Practice suggests
The Code gives practical expectations for how labels should look and sound. In summary:
- Common icon and wording. A short "AI" icon, with fuller text distinguishing content that is fully AI-generated from content that is only AI-modified.
- Placement for images and video. The label should sit where nothing covers it, for example the top corner of the frame.
- Audio. A spoken notice at the start in plain language, repeated at regular intervals for long-form or live audio.
- Text. Public-interest text generated by AI needs a disclosure unless a human has reviewed it and an editor takes responsibility for it. Regulators and lawyers read that exception narrowly, so do not assume it covers you.
- Creative and satirical work. The obligation is lighter but not removed.
- Records. Deployers should document how they apply labelling, in proportion to their size.
Check the Commission's latest guidelines before you finalise your own rules, because the detail is still being refined.
A practical workflow for a newsroom or content team
- Sort content into three groups. Fully AI-generated, AI-modified and AI-assisted. Decide where light uses such as spelling, transcription or translation sit, and write the decision down.
- Name an approver. One accountable person or desk for each content type, with a deputy.
- Set a label standard per format. Image, video, audio, live, text, and the social cut-downs that travel without context.
- Keep provenance. Record which tool, which source material and which edits produced the final asset.
- Check your tools. Ask each supplier whether outputs are marked in a machine-readable way and whether that survives export and editing.
- Train and test. Run real examples through the workflow, including freelancers and agency partners.
- Plan for corrections. Decide how a missing or wrong label is fixed and who is told.
Why this is a governance problem, not a design problem
Labels are easy to design and hard to apply consistently. Content moves through many tools, many people and many channels. Freelancers and agencies add their own tools. Archives contain older assets. A rule that lives only in an editor's head will fail under deadline pressure.
That is why the first step is the same as for any AI governance work: know what you use. Keep a current list of every generative tool in production, who owns it and what it can produce. Our guide to building an AI inventory shows how. ISO/IEC 42001 then gives the structure to add roles, human oversight, monitoring and records on top.
Does it apply to you outside the EU?
Article 50 applies to AI systems and content used in the EU, wherever the organisation is based. If you broadcast, publish or advertise to EU audiences, check your position. Many non-EU organisations are choosing a single labelling standard for all markets because it is simpler to run.
To see where your organisation stands, take the free AI Readiness Check, or look at the LeadAI Assure journey.
This article gives general information and is not legal advice or a substitute for the EU AI Act, the Code of Practice, ISO/IEC 42001:2023 or professional advice.
Sources
- Lausen, Section 50(4) of the AI Act: https://lausen.com/en/section-504-of-the-ai-act-what-organisations-must-label-as-ai-content-from-august-2026/
- Covington, Article 50 guidelines and Code of Practice: https://www.covingtonblogs.com/?p=38086
- MediaNama, EU Code of Practice on labelling AI-generated content: https://www.medianama.com/2026/06/223-eu-ai-act-deepfake-labeling-rules/
- Article 50 overview and grace period: https://artificialintelligenceact.eu/transparency-rules-article-50/