Generative AI features under Article 50
A chatbot, an image, audio or video generator, or a feature that writes text on matters of public interest falls under Art. 50 of Regulation (EU) 2024/1689 as a provider, a deployer, or both. The duties are the interaction notice under Art. 50(1), the machine-readable mark under Art. 50(2) and the labels under Art. 50(4), from the general application date In force. Regulation (EU) 2026/1744 left Art. 50 unchanged and added a marking grace period for legacy systems Deferred. This hub lists the obligation set with dates, the evidence, scenarios, guides, services and tools.
The situation
Art. 50 asks three things of such a product. Art. 50(1) requires the provider to design the interaction so that the natural person is informed that they interact with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person. Art. 50(2) requires the provider of a system that generates synthetic audio, image, video or text to mark the output in a machine-readable format and to make it detectable as artificially generated. Art. 50(4) requires the deployer to disclose a deep fake and AI-generated text published to inform the public on matters of public interest. A company that builds the feature and runs it in its own product holds the provider and the deployer duties at once. Regulation (EU) 2026/1744 did not change the content of Art. 50(1) to (6); it amended Art. 50(7) and inserted the Art. 111(4) grace period for the marking duty on systems already on the market.
The marking duty is where testing matters. Metadata, watermarks and provenance manifests are removed by ordinary re-encoding, resizing, cropping and screenshots. Art. 50(2) asks for a technical solution that is effective, interoperable, robust and reliable as far as technically feasible, so the provider has to show which mark survives which transformation and which party other than the provider can detect it. The interaction notice under Art. 50(1) is a separate check: a notice drafted for the interface has to meet Art. 50(5), which requires the information at the latest at the first interaction, in a clear and distinguishable manner, and in conformity with the applicable accessibility requirements.
A feature that is already live carries these duties from the application date, and the Art. 111(4) grace period covers only the Art. 50(2) marking duty for systems placed on the market before that date. A market surveillance authority acting under Art. 79 asks for the marking architecture, the test records and the detection solution, and a policy statement does not answer that request. The questions this hub answers:
- The feature is already live and the Article 50 duties already apply
- Nobody has measured whether the watermark survives re-encoding, cropping or a screenshot
- The disclosure text in the chat interface was drafted for the interface and never checked against Article 50(1)
- Enforcement questions from an authority would need evidence of the marking architecture, not a policy statement
- The Code of Practice on Transparency and the Commission guidelines set expectations that vendor documentation does not cover
Obligations and dates
The obligation set is one article with two dates: the Art. 50 application date, and the Art. 111(4) grace period for the marking duty on systems that generate synthetic content and were already on the market.
- In forceArticle 50 transparency obligations
- DeferredArticle 50(2) machine-readable marking for systems placed on the market before the Article 50 date
Article 50 applies from the date on the article-50-transparency entry; systems placed on the market before the Omnibus date have the marking grace period recorded on the article-50-2-legacy-marking entry.
| Obligation | Applies from | Testable | Documentation | Process | Services |
|---|---|---|---|---|---|
| Art. 50Transparency obligations for providers and deployers of certain AI systems | In force | 11 | 12 | 8 |
Also relevant, without an obligation page:
- Art. 53
Art. 50 sets no numeric threshold. Art. 50(2) sets four qualitative criteria, effective, interoperable, robust and reliable, bounded by technical feasibility, the specificities of each content type, the cost of implementation and the generally acknowledged state of the art. No harmonised standard is cited in the Official Journal for Art. 50, so no presumption of conformity is available for any technique, and no JTC 21 draft covers the article; the level of the source hierarchy in use is a Commission code and guideline: Code of Practice on Transparency of AI-Generated ContentPublished for the marking layers, the detection solution commitments and the Art. 50(4) label rules, and Commission guidelines on Article 50 transparency obligationsPublished for the scope of each duty and its exemptions. The provider sets and justifies its own targets for detection rate, false-positive tolerance and the transformation set against the context of use; the target-setting step of PR-50-MARK v1.0 records that justification. Which marking techniques count as adequate under Art. 50(2) is an unsettled point.
What an assessor or authority will ask for
An assessor or a market surveillance authority asks for records that show the marking architecture, the test outcomes and the placement of each disclosure, in this order.
- Marking architecture description: metadata, watermark and provenance layers in use, per output modality
- Marking presence and survivability test report across re-encoding, resizing, screenshot and format conversion
- Third-party detectability verification report naming the detectors and the outcomes
- Chatbot disclosure and deepfake labelling UX audit against Article 50(1) and 50(4)
- Remediation list with the re-test criteria for each finding
Three artifact families on deliverables answer that list. The test report family carries the marking presence and survivability results per modality and per transformation, and the third-party detectability results naming the detectors and the outcomes. The assessment memo family carries the Article 50 applicability and exemption determination record and the interaction disclosure and label audit record, each with its article reference. The findings register carries every gap with the re-test criterion that closes it. Each family is specified with a synthetic sample, and the protocol run record states the version of the protocol under which each result was produced.
Scenarios
Each scenario page states a situation this product type meets, the sequence of checks, the evidence that results, and the consequence of inaction with its article reference.
- Are your Article 50 transparency obligations for AI content in order: evidence for an enforcement query
- Our chatbot went live: the European AI Act chatbot checklist under Article 50
Guides
Each guide gives a command, a configuration or a synthetic example for one Art. 50 test, with a threshold or pitfall and its source.
- Transparency obligations under Article 50: implementing each of the four
- Verify content credentials: C2PA manifest validation step by step
- Code of practice on transparency of AI-generated content: what it adds to Article 50
- EU AI Act watermark survivability: testing a watermark against re-encoding
Services
Two services test against Art. 50 for this product type, within the limits stated on what we cannot do: the Art. 50 testing service runs the marking, survival, detection and disclosure checks, and the readiness assessment assigns each function to its Art. 50 paragraphs and checks each claimed exemption.
- Article 50 transparency and marking/watermark testingnow, past legacy graceIn force
- Classification and readiness assessmentnowIn force
Protocols
Each protocol is versioned and published in full on methodology, with a citation on every step.
- PR-50-MARK v1.0Article 50 marking and detection protocol
- PR-CLS v1.0Classification and role determination protocol
Tools and templates
The tools run in the browser, send nothing, and follow the logic published on the tool methodology page; the document outlines are on templates.
Questions
What is Article 50 of the EU AI Act?
Art. 50 of Regulation (EU) 2024/1689 sets transparency duties for certain AI systems. Art. 50(1) requires a provider to inform natural persons that they interact with an AI system. Art. 50(2) requires a provider of a system that generates synthetic audio, image, video or text to mark the output in a machine-readable format and to make it detectable. Art. 50(4) requires a deployer to disclose deep fakes and AI-generated public-interest text. Art. 50(5) requires the information to be clear, distinguishable, timely and accessible.
Does the AI Act give a definition of generative AI?
Regulation (EU) 2024/1689 has no defined term for generative AI. Art. 50(2) describes the systems it covers as AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content. Art. 3(1) defines an AI system, Art. 3(60) defines a deep fake, Art. 3(63) defines a general-purpose AI model and Art. 3(66) defines a general-purpose AI system. The Art. 50(2) marking duty attaches to the generating function, whatever the product calls it.
What does Article 50(2) of the AI Act require?
Art. 50(2) requires a provider of an AI system that generates synthetic audio, image, video or text to mark the output in a machine-readable format and to make it detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible. The duty does not apply to an assistive function for standard editing. Art. 111(4), inserted by Regulation (EU) 2026/1744, gives systems already on the market a grace period for the Art. 50(2) marking duty; the article-50-2-legacy-marking entry carries the date.
What do the Commission guidelines on Article 50 of the AI Act cover?
The Commission guidelines on Article 50 transparency obligations, a published guideline, give the Commission's reading of each Art. 50 duty. The guidelines cover the reasonably well-informed person test for the Art. 50(1) exemption, the reading of deep fake, examples of the assistive-editing and no-substantial-alteration exemptions under Art. 50(2), the two cumulative conditions for the Art. 50(4) editorial-control exemption, and the timing and accessibility reading of Art. 50(5). The guidelines are not binding law; the consolidated text on EUR-Lex governs.
Every inquiry receives a reply within 24 hours. Scope and availability are discussed per request.