Article 50 transparency obligations: deepfake disclosure and synthetic content labelling
As of 18 November 2026 · checked against Regulation (EU) 2024/1689, Regulation (EU) 2026/1744 and the guidelines C/2026/6005 · Deutsch
Information only, not legal advice. Article 50 transparency duties apply across providers and deployers. Confirm any concrete labelling decision with qualified counsel.
Article 50 of Regulation (EU) 2024/1689 carries the AI Act's user-facing transparency rules. It does not depend on Annex III classification: it applies to any system that interacts with humans, generates synthetic content, performs emotion recognition or biometric categorisation, or produces deepfakes. The obligations are split between providers (who must build the technical capability) and deployers (who must actually disclose). This page summarises both sides and what the Commission guidelines of November 2026 say about machine-readable marking.
The four trigger cases
| Case | Who discloses | What must be disclosed |
|---|---|---|
| AI system interacts with humans | Provider builds, deployer ensures | The fact that the user is interacting with an AI system |
| Emotion recognition or biometric categorisation | Deployer | Operation of the system to the natural persons exposed |
| Synthetic image, audio, video, or text | Provider marks output | Machine-readable marker that the content is artificially generated or manipulated |
| Deepfake or AI-generated text on matters of public interest | Deployer | Visible disclosure that the content is generated or manipulated |
Provider duty: machine-readable marking and detection
Article 50(2) requires providers of systems that generate synthetic audio, image, video or text to mark the outputs in a machine-readable format and make them detectable as artificially generated or manipulated. The technical solutions must be effective, interoperable, robust and reliable as far as technically feasible, taking into account the generally acknowledged state of the art. The duty does not apply where the system only performs an assistive function for standard editing or does not substantially alter the input. What the guidelines C/2026/6005 add:
- Techniques. One technique or a combination is allowed, as long as the overall solution is machine-readable and meets the four requirements. Recital 133 names watermarks, metadata identifications, cryptographic provenance methods, logging and fingerprints. A full provenance chain is not required.
- Detection. Providers must make a means of detection available to the people exposed to the content, with human-readable results. Publicly available industry standard detection is the target; own or shared detectors are a stopgap until such standards exist.
- Upstream marking. A provider may rely on marking implemented by an upstream model provider or a third party, but stays responsible for showing compliance.
- Narrow exceptions. Less robust metadata marking can be enough for output that stays in a closed product environment. Strictly technical business-to-business output seen only by a small number of professionals can need no marking, if all the conditions in the guidelines are met.
The guidelines name no specific standard. An earlier version of this page said the Commission would recognise C2PA through an implementing act and gave per-modality recipes. That is not in the guidelines or the Regulation, so it has been removed. Providers may rely on a code of practice assessed as adequate under Article 50(7).
Deployer duty: visible disclosure for deepfakes
Deployers of systems that generate or manipulate image, audio, or video resembling real persons, places, or events (deepfakes) must disclose that the content is artificially generated or manipulated. Two practical carve-outs exist:
- Law-enforcement use. Disclosure can be withheld where authorised by law to detect, prevent, or prosecute criminal offences.
- Evidently artistic, creative, satirical, or fictional works. Disclosure is still required, but limited to an appropriate form that does not hamper the display or enjoyment of the work. The guidelines treat the form as a case-by-case assessment.
For AI-generated text published on matters of public interest, the disclosure obligation kicks in unless the content went through human editorial review and a person or organisation takes editorial responsibility.
Implementation pattern: labelling pipeline
A clean pipeline separates marker generation (provider responsibility) from disclosure rendering (deployer responsibility). Example for a generative image API:
Illustrative request shape, not an API of ours and not a format the Regulation prescribes:
POST /v1/images/generate
{
"prompt": "...",
"labelling": {
"c2pa": true,
"watermark": "stable-signature-v2",
"deployer_disclosure": {
"render": "corner-badge",
"text": "AI-generated"
}
}
}
→ response includes:
- image bytes with embedded C2PA + watermark (Article 50(2))
- rendered visible disclosure overlay (Article 50(4))
- signed manifest URL for downstream verifiers (audit trail) What enforcement is likely to look like
- Any natural or legal person with grounds to consider that the Regulation has been infringed can complain to a market surveillance authority (Article 85).
- Penalties under Article 99(4) reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher, for breach of Article 50.
- The transparency obligations have applied since 2 August 2026. The Digital Omnibus did not postpone them. Providers of generative systems that were on the market before that date have until 2 December 2026 to comply with the marking duty of Article 50(2) (Article 111(4)). All dates and sources are in the deadline timeline.
- The Commission guidelines on Article 50 were published in the Official Journal on 12 November 2026 (OJ C, C/2026/6005). They are not binding.
What we ship as events
AI Act Radar emits events on the transparency topic when the Commission, the AI Office, standardisation bodies or national authorities publish guidance on Article 50, including codes of practice and their assessment under Article 50(7). Subscribe with topic filter transparency.
Sources
- Regulation (EU) 2024/1689, Articles 50, 85, 99 and recitals 133 and 134: EUR-Lex
- Regulation (EU) 2026/1744, new Article 111(4) and amended Article 50(7): EUR-Lex
- Commission guidelines on Article 50, OJ C, C/2026/6005, 12 November 2026: EUR-Lex
→ See live transparency updates: https://aiactradar.com
→ Subscribe to webhook: https://docs.aiactradar.com/quickstart