Under Article 50 of the EU AI Act, providers must machine-mark synthetic audio, images, video, and text, and deployers must clearly disclose deepfakes and AI-written content on matters of public interest — with a marking grace period running into December 2026.
2026-07-31 · by Moe Ameen
On August 2, 2026, the transparency obligations in Article 50 of the EU AI Act begin to apply. Two of them land directly on anyone who makes or publishes content. The first sits with providers of generative AI systems: their tools must mark synthetic output so it is machine-readable and detectable as artificially generated or manipulated. The second sits with deployers — the people and businesses using AI to make content — who must clearly disclose deepfakes and certain AI-written text.
The provider marking duty (Article 50(2)) covers synthetic audio, images, video, and text. The mark has to be robust and detectable in a machine-readable way, which in practice points toward watermarks or embedded metadata rather than a caption a user can delete. The Act carves out exceptions: systems that perform an assistive editing function, or that do not substantially alter the input data or its meaning, are not caught by the marking requirement.
The deployer disclosure duty (Article 50(4)) is the one most creators will feel. Deepfakes — AI-generated or manipulated image, audio, or video that resembles real people, objects, places, or events and could falsely appear authentic — must be clearly labeled as artificially generated. AI-generated or manipulated text published to inform the public on matters of public interest must also be disclosed, unless it went through human review with a person or organization holding editorial responsibility. Disclosure can be less obtrusive for work that is evidently artistic, creative, satirical, or fictional.
A few timing and enforcement details matter. The rules apply from August 2, 2026, with a grace period running into December 2026 for generative AI systems already placed on the market before that date, and content that was both generated and published before August 2, 2026 does not have to be labeled retroactively. Non-compliance can draw fines up to €15 million or 3% of worldwide annual turnover, whichever is higher. The European Commission has published draft guidelines on Article 50 and is finalizing a Code of Practice on marking and labeling, so some practical detail is still settling — treat the technical specifics as evolving rather than fixed.
The part of Article 50 that touches your daily workflow is the deployer disclosure duty — and the review step that can lift it for text. Kompozy is a full AI content generation and multi-platform publishing engine with a per-post review pipeline built in: every generated asset (persona and avatar video, Clipped Shorts, Carousels, Photo Posts, Quote Graphics, Blog Articles, Newsletters, and more) passes an approval gate before it schedules. That gate is a natural home for the human editorial review the AI Act treats as lifting the disclosure obligation on AI-written content, and a natural place to attach a standard "AI-generated" disclosure line to your captions or brand templates once — instead of remembering it per platform.
Because Kompozy fans one source out to the eight primary social platforms plus blog and email, you set your labeling convention in a single place and it rides along to Instagram, Facebook, TikTok, YouTube, LinkedIn, X, Pinterest, Threads, your blog, and your newsletter. The Persona Brief governs voice; your caption template can carry the disclosure text consistently across every output. Kompozy does not itself stamp the machine-readable watermark that Article 50(2) puts on the generation model's provider, and it is not legal advice — but it gives you one on-brand pipeline where a real review step and consistent disclosure are the default, not something bolted on at each platform after the fact. Read the Commission's Article 50 guidelines and check with counsel for your specific situation.
The transparency obligations in Article 50 of the EU AI Act begin to apply on August 2, 2026. A grace period runs into December 2026 for generative AI systems already on the market before that date, and content that was both generated and published before August 2, 2026 does not have to be labeled retroactively.
Both, in different ways. The provider of the generative AI system must apply a machine-readable mark so output is detectable as artificially generated (Article 50(2)). The deployer — the person or business using it — must clearly disclose deepfakes and AI-generated or manipulated text on matters of public interest (Article 50(4)), unless that text went through human review with editorial responsibility.
No. The rules target synthetic and manipulated content. Assistive editing functions and changes that do not substantially alter the input or its meaning are exempted, and disclosure can be less obtrusive for clearly artistic, creative, satirical, or fictional work. Fully AI-generated media and realistic deepfakes carry the clearest duty.
Non-compliance with the transparency obligations can draw fines up to €15 million or 3% of a company's worldwide annual turnover, whichever is higher. Enforcement runs through national market surveillance authorities alongside the EU AI Office. The Commission has published draft guidelines and is finalizing a Code of Practice on marking and labeling, so some practical detail is still being settled.