EU AI Act, Article 50 (transparency obligations for AI-generated content)

What it requires

Article 50 collects the AI Act's transparency duties. Two of them bear on synthetic media and run on different actors: Article 50(2) requires providers of generative AI systems to mark synthetic output in a machine-readable form so that it can be detected as artificially generated, and Article 50(4) requires deployers to disclose deep fakes to the people who see them. The Commission's page on the EU labelling icons states that not all AI-generated or manipulated content needs to be labelled: Article 50(4) reaches deep fakes, defined as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic, and AI-generated or manipulated text published to inform the public on matters of public interest without human review or editorial responsibility. Where deep fake content forms part of an evidently artistic, creative, satirical, fictional or analogous work, the Commission states that the transparency obligation is limited to disclosure in an appropriate manner that does not hamper the display or enjoyment of the work.

Why it bears on this field

Vertical-drama services sit in the deployer category that the Code of Practice calls Section 2, and a large share of the format's output is machine-made: TikTok is reported to claim that 40% of short-drama viewing time on its platform is now AI content, a self-reported and unaudited figure carried in DataEye's first-half 2026 overseas report, and Kunlun Tech says its DramaWave app is more than 80% AI-generated. On the face of the Commission's definition, a wholly AI-generated series with invented characters, invented places and invented events does not resemble existing persons or events and is therefore not a deep fake; that reading is supported by the text the Commission has published but is not stated by it, the guidelines are not binding, and only the Court of Justice can interpret Article 50 authoritatively. Where a series uses a real person's likeness, such as a celebrity cameo, a de-aged performer or a synthetic voice of a living actor, the deep-fake duty applies. The human-review carve-out covers text only, not video, and signatories of the Code are bound by its Section 2 placement specifications whatever the statutory floor turns out to be.

Open questions

• Does a wholly fictional, wholly AI-generated series fall outside the deep-fake definition in Article 50(4)? The Commission's published text supports that reading without stating it.
• What counts as disclosure in an appropriate manner that does not hamper the display or enjoyment of the work, for a fictional series?
• How is the requirement that a label remain visible when content is reshared or downloaded to be met on services whose distribution runs on resharing?
• Which national authority is competent for Article 50 in Germany? See the entry on German supervision of AI Act transparency.

Timeline

• 2026-06-10The Commission publishes the Code of Practice on Transparency of AI-generated Content, a voluntary instrument with a provider section (Section 1) and a deployer section (Section 2). Source (opens in a new tab)
• 2026-07-08The Commission concludes that the Code adequately covers the obligations in Article 50(2), (4) and (5). The AI Board adopts its adequacy assessment the following day and both documents are published on 9 July. The Commission also states that adherence to the code does not constitute conclusive evidence of compliance, and that non-signatories must demonstrate compliance through alternative equivalently adequate means. Source (opens in a new tab)
• 2026-07-20Guidelines on transparency obligations for providers and deployers of AI systems are published, together with the Communication approving their content. They are not binding, though the AI Office and national authorities are expected to follow them. Source (opens in a new tab)
• 2026-07-20Date carried by the Commission's page of EU icons for labelling AI-generated content. Three icons — Basic, Fully AI-Generated and Partially AI-Modified — are offered in four variations as SVG and PNG, free to use without attribution. The Commission states that use of the icons by non-signatories of the Code should not be construed as signalling adherence to it, and that using them does not by itself establish compliance. Its placement rules require a label to be perceivable and distinguishable at the latest at first exposure and to remain visible when content is reshared or downloaded. Commission user-testing found that performance improved across all measures when the basic icon was accompanied by a text label. Source (opens in a new tab)
• 2026-07-2218:00 CEST cut-off for inclusion in the first published list of signatories to the Code. Signing remains possible afterwards; the legally binding date is 2 August 2026. Source (opens in a new tab)
• 2026-07-27Regulation (EU) 2026/1744, the Digital Omnibus on AI, dated 8 July 2026 and published in the Official Journal on 24 July, enters into force. It does not move the Article 50 application date. It does touch Article 50 twice: it replaces Article 50(7) and inserts a transitional provision at Article 111(4). It defers stand-alone Annex III high-risk obligations to 2 December 2027.
• 2026-07-31The Commission publishes the signatory list: about 190 organisations, of which 83 signed Section 1 (providers) and 152 Section 2 (deployers). The two figures overlap, because some organisations signed both sections. The Commission says about half the signatories are small and recent companies. Section 1 signatories include Google, Meta, Microsoft, Mistral, OpenAI, Anthropic, Cohere, Aleph Alpha, Black Forest Labs and Synthesia; Section 2 signatories include Getty Images and Studio Funk. Two signatory task forces are due to launch in September 2026. Source (opens in a new tab)
• 2026-08-02The Article 50 transparency obligations begin to apply. Breaches carry fines of up to €15m or up to 3% of total worldwide annual turnover under Article 99 of the AI Act; the higher €35m ceiling applies to the prohibited-practices tier, per Medien.Bayern. Source (opens in a new tab)
• 2026-12-02Deadline for machine-readable marking under Article 50(2) for generative systems placed on the market before 2 August 2026, under the transitional inserted by Regulation (EU) 2026/1744. Outputs generated and already made available before 2 August 2026 do not have to be marked retroactively. Article 50(4) is unaffected and has applied since 2 August 2026. Source (opens in a new tab)