Art. 50: Transparency obligations for certain AI systems under the EU AI Act

Art. 50 requires providers to inform natural persons of AI interaction and to mark synthetic audio, image, video and text in a machine-readable format. Deployers disclose emotion recognition, biometric categorisation and deep fakes. In this mapping, 11 elements are testable, 12 are documentation and 8 are process. Regulation (EU) 2026/1744 did not move the application date 2026-08-02In force. Art. 111(4) gives systems already on the market a grace period for the Art. 50(2) marking duty 2026-12-02Deferred. The adequate Art. 50(2) marking technique is an unsettled point; the page shows both readings.

Where this applies

2026-08-02In forceArticle 50 transparency obligations
2026-12-02DeferredArticle 50(2) machine-readable marking for systems placed on the market before the Article 50 date
2026-07-27In forceDigital Omnibus on AI, entry into force

What the text requires

Art. 50(2) holds most of the testable elements: mark presence per modality, detectability by a third party, survival under ordinary transformations, readability by an independent tool, and detection reliability against a justified target. Each exemption in Art. 50(1) to (4) is a documentation element, because the exemption rests on a recorded determination.

What the text requires and how each element is verified
Requires	Testable	Documentation	Process
Art. 50(1)Providers must design and develop AI systems intended to interact directly with natural persons so that the natural persons concerned are informed that they are interacting with an AI system. This does not apply where the interaction is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use. This does not apply to AI systems authorised by law to detect, prevent, investigate or prosecute criminal offences, subject to appropriate safeguards for the rights and freedoms of third parties, unless those systems are available for the public to report a criminal offence. Art. 50(2)Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content must ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. Providers must ensure that their technical solutions are effective, interoperable, robust and reliable as far as this is technically feasible. Providers take into account the specificities and limitations of the types of content, the costs of implementation and the generally acknowledged state of the art, as may be reflected in relevant technical standards. This does not apply to the extent the AI system performs an assistive function for standard editing or does not substantially alter the input data provided by the deployer or the semantics of that data. This does not apply where the system is authorised by law to detect, prevent, investigate or prosecute criminal offences. Regulation (EU) 2026/1744 did not change this paragraph; Art. 111(4) gives systems placed on the market before the Art. 50 application date a grace period for this duty. Art. 50(3)Deployers of an emotion recognition system or a biometric categorisation system must inform the natural persons exposed to the system of its operation. Deployers must process the personal data in accordance with Regulation (EU) 2016/679, Regulation (EU) 2018/1725 and Directive (EU) 2016/680, as applicable. A system that is ancillary to another commercial service and strictly necessary for objective technical reasons is not a biometric categorisation system under the definition in Art. 3(40), so Art. 50(3) does not reach it. This does not apply to systems permitted by law to detect, prevent or investigate criminal offences, subject to appropriate safeguards and in accordance with Union law. Art. 50(4)Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake must disclose that the content has been artificially generated or manipulated. This does not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences. Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the duty is limited to disclosure of the existence of the generated or manipulated content in a manner that does not hamper the display or enjoyment of the work. Deployers of an AI system that generates or manipulates text published with the purpose of informing the public on matters of public interest must disclose that the text has been artificially generated or manipulated. This does not apply where the use is authorised by law for criminal-offence purposes, or where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication. Art. 50(5)The information under paragraphs 1 to 4 must be given to the natural persons concerned in a clear and distinguishable manner. The information must be given at the latest at the time of the first interaction or exposure. The information must conform to the applicable accessibility requirements. Art. 50(6)Paragraphs 1 to 4 do not affect the requirements and obligations in Chapter III. Paragraphs 1 to 4 are without prejudice to other transparency obligations laid down in Union or national law for deployers of AI systems. Art. 50(7)The Commission encourages and facilitates the drawing up of codes of practice at Union level for the effective implementation of the obligations on the detection, marking and labelling of artificially generated or manipulated content. The Commission, taking utmost account of the opinion of the Board, assesses whether adherence to those codes is adequate for the obligations in paragraphs 2 and 4, under the procedure in Art. 56(6). If the Commission deems a code inadequate, it may adopt an implementing act with common rules for the implementation of those obligations under the examination procedure in Art. 98(2). Regulation (EU) 2026/1744 amended this paragraph: the Commission, not the AI Office, now leads the codes, the paragraph names marking alongside detection and labelling, and adequacy is assessed in a single step with the opinion of the Board. Art. 111(4)Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content and were placed on the market before the Art. 50 application date must take the necessary steps to meet Art. 50(2) by the end of the grace period stated in this paragraph. Regulation (EU) 2026/1744 inserted this paragraph.	Each function of the system is assigned to the Art. 50 paragraphs that apply to it (interaction, generation, emotion recognition or biometric categorisation, deep fake, public-interest text), and each claimed exemption under Art. 50(1), (2), (3) or (4) is checked against the observed behaviour of the function. A natural person who starts an interaction sees or hears that they interact with an AI system at the latest at the first interaction, in a clear and distinguishable manner, without opening a separate page or document (Art. 50(1) with Art. 50(5)). Every synthetic output of each modality (audio, image, video, text) carries a machine-readable mark on generation, verified on a sample of outputs per modality and per output path (Art. 50(2)). Each marked output is detectable as artificially generated or manipulated by a detection solution that a party other than the provider can run, verified by running the detection on marked samples (Art. 50(2), detectability). The mark survives a defined set of ordinary transformations for the modality (re-encoding, resizing, cropping, screenshot, format conversion, transcoding, copy and paste for text), verified per transformation with the survival result recorded (Art. 50(2), robustness). The mark is readable by tools other than the provider's own, such as a provenance manifest validator or a shared detection interface, verified against at least one independent reader (Art. 50(2), interoperability). The detection solution returns the correct result on marked outputs and on non-generated control samples, with the detection rate and the false-positive rate recorded against the target the provider justified (Art. 50(2), reliability and effectiveness). A natural person exposed to an emotion recognition or biometric categorisation system receives the notice of its operation at the latest at the first exposure, verified on the deployment interface or premises (Art. 50(3) with Art. 50(5)). Deep fake content shown to a natural person carries a disclosure that is perceivable at the latest at the first exposure, embedded so that ordinary display does not hide it, verified on each distribution surface (Art. 50(4) with Art. 50(5)). Published AI-generated text on matters of public interest carries a disclosure at the point of publication, unless the editorial-control exemption is recorded for that publication, verified on the published surface (Art. 50(4) with Art. 50(5)). Each disclosure under Art. 50(1), (3) and (4) conforms to the applicable accessibility requirements, verified with the accessibility checks the applicable requirement prescribes (Art. 50(5)).	Record of the assessment that an interaction is obvious to a reasonably well-informed, observant and circumspect person, for each function where the provider relies on the Art. 50(1) exemption. Marking architecture description per output modality: the metadata, watermark and provenance layers in use, the point in the pipeline where each is applied, and the detection solution that reads each layer (Art. 50(2)). Justification of the technical solution against the four Art. 50(2) criteria (effective, interoperable, robust, reliable), stating the technical feasibility limits, the specificities of each content type, the cost of implementation and the state of the art relied on. Exemption determination record for each generating function claimed to perform an assistive function for standard editing or not to substantially alter the input data or its semantics (Art. 50(2)). Detection solution description available to third parties: how a party other than the provider detects the mark, on what terms, and for which modalities (Art. 50(2), detectable). Data protection record for the personal data processed by an emotion recognition or biometric categorisation system, and the record of the Art. 3(40) ancillary-and-strictly-necessary determination where the system is treated as outside the definition of a biometric categorisation system. Editorial responsibility record naming the natural or legal person who holds editorial responsibility, and a description of the human review or editorial control measures, for each publication where the Art. 50(4) exemption is relied on. Artistic-work determination record for each deep fake use where the limited disclosure under Art. 50(4) is relied on. Law-enforcement authorisation record for any function that relies on the criminal-offence exception in Art. 50(1), (2), (3) or (4). Obligation register entry that lists the Chapter III requirements and the other Union or national transparency duties that apply alongside Art. 50 (Art. 50(6)). Record of adherence to the Code of Practice on Transparency of AI-Generated Content, or of the alternative means used to show conformity with Art. 50(2) and (4) (Art. 50(7)). Grace-period record for a system placed on the market before the Art. 50 application date: the steps taken to meet Art. 50(2) and the date each step completed (Art. 111(4)).	Re-run the marking presence, survivability and detectability tests after each change to the model, the output encoder, the distribution surface or the marking library (Art. 50(2)). Review the marking technique against the generally acknowledged state of the art and relevant technical standards at a set interval, and record the review outcome (Art. 50(2)). Keep the detection solution available and operating for the outputs of every version of the system that remains in use (Art. 50(2), detectable). Process the personal data of persons exposed to an emotion recognition or biometric categorisation system under the applicable data protection law throughout the use of the system (Art. 50(3)). Operate the human review or editorial control workflow for public-interest text, and keep the named editorial responsibility current, wherever the Art. 50(4) exemption is relied on. Operate the labelling workflow that attaches the deep fake or text disclosure before publication on each distribution surface (Art. 50(4)). Monitor the Commission's adequacy assessment of the Code of Practice and any implementing act with common rules, and update the marking and labelling measures when either changes (Art. 50(7)). Complete the steps to bring a system placed on the market before the Art. 50 application date into conformity with Art. 50(2) within the Art. 111(4) grace period.

Metrics and where thresholds come from

Threshold sources, in order of precedence

A harmonised standard cited in the Official Journal (none as of the anchor date)
A JTC 21 draft standard (prEN), with its stage
An ISO/IEC standard or technical report, with its stage
A Commission guideline or code of practice, with its date from the standards data
NIST AI RMF or a sector framework
The state of the art, which the provider justifies and documents

Art. 50 sets no numeric threshold. Art. 50(2) sets four qualitative criteria, effective, interoperable, robust and reliable, bounded by technical feasibility, the specificities and limitations of each content type, the costs of implementation and the generally acknowledged state of the art. No harmonised standard is cited in the Official Journal for Art. 50 and no JTC 21 draft covers it, so the hierarchy level in use today is a Commission code of practice and guideline: the Code of Practice on Transparency of AI-Generated Content and the Commission guidelines on Article 50, with the C2PA specification as a technique reference. Where the Code states a quantity, the record cites the Code and the condition it attaches; no quantity is attributed to the Act. The provider sets and justifies its own targets for detection rate, false-positive tolerance and the transformation set against the context of use; the PR-50-MARK target-setting step records the justification.

Metric sources and their stage
Source and stage	Title	What it provides here	Verified on
`Code of Practice on Transparency of AI-Generated Content`Published	Code of Practice on Transparency of AI-Generated Content (marking and labelling under Art. 50(2) and 50(4))	operational measures for the Art. 50(2) marking layers (digitally signed and time-stamped metadata, imperceptible watermark, with fingerprinting and logging as supplementary measures only), a multi-layer expectation where a single technique cannot meet the criteria on its own, detection solution availability and interoperability commitments, and the Art. 50(4) label placement rules and EU icons	Verified on 18 September 2026European Commission, Code of Practice on Transparency of AI-Generated Content
`Commission guidelines on Article 50 transparency obligations`Published	Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act	the scope of each Art. 50 duty: the reasonably well-informed person test for the Art. 50(1) exemption, the reading of deep fake, the assistive-editing and no-substantial-alteration exemption examples, the two cumulative conditions for the Art. 50(4) editorial-control exemption, and the timing and accessibility reading of Art. 50(5)	Verified on 18 September 2026European Commission, Guidelines on transparency obligations for providers and deployers of AI systems
`C2PA Content Credentials specification`Published	C2PA Technical Specification (Content Credentials)	the provenance manifest structure, signature validation rules and manifest-binding checks used to verify the metadata and provenance layer for presence, validity and survival through edits	Verified on 18 September 2026C2PA Specifications 2.1 (version selector also lists 2.2, 2.3 and 2.4)

Procedures

Each protocol is versioned; its page states the derivation from the standards above. PR-50-MARK v1.0 covers the marking, survival and detection tests under Art. 50(2). PR-CLS v1.0 covers the assignment of each function to the Art. 50 paragraphs and the check of each claimed exemption.

PR-50-MARK v1.0Article 50 marking and detection protocolv1.0
PR-CLS v1.0Classification and role determination protocolv1.0

What an assessor asks for

Article 50 applicability and exemption determination record (PR-CLS output)
Marking architecture description per output modality
Technical solution justification against the Art. 50(2) criteria
Marking presence and survivability test report (PR-50-MARK output)
Third-party detectability verification report (PR-50-MARK output)
Interaction disclosure and label audit record (PR-50-MARK output)
Editorial responsibility and human review measures record (Art. 50(4))
Code of Practice adherence or alternative-means record (Art. 50(7))

Common findings

Metadata is the only marking layer, and a screenshot or a re-encode removes it.
The watermark detector runs only inside the provider's own service, so no other party can detect the mark.
The interaction disclosure sits in the terms of service or a help page rather than in the interface at the first interaction.
The deep fake label is an overlay on the player, and a crop, a download or a re-share removes it.
The assistive-editing exemption is claimed for a function that replaces most of the input content.
The technical solution is chosen without a written justification against the four Art. 50(2) criteria.

Questions

What does Article 50 machine-readable marking require?

Art. 50(2) requires providers of systems that generate synthetic audio, image, video or text to mark the outputs in a machine-readable format. The outputs must also be detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible. The Act names no technique and no number. The Code of Practice on Transparency of AI-Generated Content names signed metadata and imperceptible watermarks as marking layers and treats fingerprinting and logging as supplementary. Which techniques count as adequate is an unsettled point; the page shows both readings.

What does the Code of Practice on Transparency of AI-Generated Content add to Article 50?

The Code is a voluntary instrument under Art. 50(7). Its first section covers provider marking and detection under Art. 50(2). Its second section covers deployer labelling under Art. 50(4). It expects more than one marking layer where a single technique cannot meet the four criteria on its own. It commits signatories to make a detection solution available. It supplies EU icons for deep fake and AI-modified content labels. The Commission assesses whether adherence to the Code is adequate, taking utmost account of the opinion of the Board. A provider that does not adhere shows conformity with Art. 50(2) and (4) by other means.

What is the EU AI Act Article 50 deadline?

Art. 50 applies from the general application date set in Art. 113; the article-50-transparency entry carries the date. Regulation (EU) 2026/1744 did not move that date. It added Art. 111(4), a grace period for the Art. 50(2) marking duty for systems that generate synthetic content and were placed on the market before the application date; the article-50-2-legacy-marking entry carries that date. The grace period covers only Art. 50(2). Disclosure duties under Art. 50(1), (3) and (4) have no grace period.

What does the AI Act require of chatbots?

Art. 50(1) requires a provider to design a system that interacts directly with natural persons so that those persons are informed that they interact with an AI system. The information must be clear and distinguishable and given at the latest at the first interaction under Art. 50(5). The duty does not apply where the interaction is obvious to a reasonably well-informed, observant and circumspect person. The Commission guidelines on Article 50 read the test as an average-person test that the provider records. A chatbot that also generates text may carry the Art. 50(2) marking duty for that text.

What does AI Act Article 50(4) require?

Art. 50(4) is a deployer duty. A deployer that generates or manipulates a deep fake must disclose that the content is artificially generated or manipulated. For an evidently artistic, creative, satirical, fictional or analogous work, the duty is limited to a disclosure that does not hamper the work. A deployer that publishes AI-generated text to inform the public on matters of public interest must disclose that the text is artificially generated or manipulated. That text duty does not apply where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility. The Commission guidelines on Article 50 treat these as two cumulative conditions and do not count spell-checking as human review.

Did the Omnibus delay the EU AI Act transparency obligations?

No. Regulation (EU) 2026/1744 did not change the content of Art. 50(1) to (6) and did not move the Art. 50 application date. It amended Art. 50(7) so that the Commission leads the codes of practice and assesses their adequacy with the opinion of the Board. It inserted Art. 111(4), a grace period for the Art. 50(2) marking duty for systems that generate synthetic content and were already on the market. The article-50-transparency and article-50-2-legacy-marking entries carry the dates.

How we test this

We test against Art. 50(1) to (5): the interaction notice on the first contact, the machine-readable mark on a sample of outputs per modality, its survival under a defined transformation set, its detection by an independent reader, and the disclosure on each distribution surface. We prepare the evidence as the PR-50-MARK outputs and the determination records listed above, within the limits stated on what we cannot do.

Classification and readiness assessmentPR-CLS
Article 50 transparency and marking/watermark testingPR-50-MARK

What we cannot do

We do not perform conformity assessments where a notified body is required.
We do not issue CE marking or the EU Declaration of Conformity.
We are not a notified body.
We do not give legal opinions on classification or liability.
We test against requirements, prepare evidence, and structure documentation; compliance decisions remain with the provider or deployer, their notified body, and their counsel.

Read the full statement