Transparency Obligations under Article 50 AI Act: Chatbots, Synthetic Content and Deepfakes

Article 50 of the EU AI Act sets transparency obligations for providers and deployers of AI systems that interact with people, generate synthetic content, recognise emotions or categorise people biometrically, or produce deep fakes. Its seven paragraphs form the sole article of Chapter IV and apply from 2 August 2026, with one transition left: providers of generative AI systems placed on the market before that date meet the marking duty of Article 50(2) by 2 December 2026 (Article 111(4)). Breaches fall under the fine tier of Article 99(4).

The AI Act is Regulation (EU) 2024/1689, in force since 1 August 2024 and directly applicable in every Member State, as amended by Regulation (EU) 2026/1744 (the Digital Omnibus on AI), in force since 27 July 2026. Article 50 transparency forms one tier of the EU AI Act, alongside the prohibited practices and the high-risk rules. Its application date is also the general date of application of the act (Article 113, second paragraph).

Compiled by the AIActGuide.net Editorial Team from the Official Journal text; how we check every statement: methodology.

What Are the Transparency Obligations under Article 50?

The transparency obligations under Article 50 are information duties, in paragraphs 1 to 4, that tell natural persons when they deal with an AI system or with artificially generated or manipulated content, given in the manner of paragraph 5. Paragraphs 1 to 6 still read as adopted in 2024, while Article 1, point (20), of Regulation (EU) 2026/1744 replaced paragraph 7 on codes of practice. Recitals 132 to 137 explain the article.

Marking versus Labelling: How Article 50 Splits Duties between Providers and Deployers

Marking is the provider duty of Article 50(2), machine-readable and detectable, while labelling is the deployer duty of Article 50(4), visible to the person exposed. A mark lets software detect that an output is artificially generated or manipulated, and a label tells a natural person the same fact. The two duties differ in five attributes:

Commission terminology, not a legal term of Article 3.

Paragraph 117 of the Commission guidelines adds that deployers cannot rely on the provider's machine-readable marking. The act itself speaks of marking in paragraph 2 and of disclosure in paragraph 4.

Article 50 within the Four Risk Tiers: From Prohibited Practices to Minimal Risk

Article 50 attaches information duties to four kinds of AI system function, whatever their risk tier, and neither prohibits nor classifies a system. The Commission names four tiers in paragraph 2 of its guidelines C(2025) 5052: unacceptable risk (prohibited AI practices), high risk, transparency risk (often called limited risk) and minimal or no risk, Commission vocabulary rather than defined terms of the act. Transparency duties differ from the prohibited AI practices of Article 5, whose points (a), (b) and (c) to (h) apply since 2 February 2025 and whose points (ba) and (bb) and paragraphs 1a and 1b apply from 2 December 2026.

The EU AI Act risk categories overlap: a system that is high-risk under Article 6 carries its Article 50 duties as well, because Article 50(6) leaves Chapter III untouched.

The Transparency Section of This Guide: Four Pages and What Each Answers

Four Article 50 topics are covered in depth in the transparency section of this guide: the chatbot disclosure of Article 50(1), the marking duty of Article 50(2), the deployer labelling of Article 50(4) and the transparency code under Article 50(7):

Article 50(3), 50(5) and 50(6), the Commission guidelines, the status of the code and enforcement are answered in the sections of this hub.

Article 50 Duties by Paragraph: Provider and Deployer Obligations

Article 50 assigns paragraphs 1 and 2 to providers and paragraphs 3 and 4 to deployers, while paragraphs 5 to 7 govern the manner, the relation to other law and the codes:

Article 50 of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744.

Each paragraph binds one of two roles defined in Article 3, provider or deployer, rather than a type of company. The provider vs deployer roles decide the paragraph, and one company can hold both for the same system. A natural person using a generator in a purely personal non-professional activity is not a deployer (Article 2(10); guidelines, paragraphs 19 to 20), while the provider still marks the output.

Paragraphs 1 and 2: What Providers Build into the System

Providers design interactive AI systems so that natural persons are informed they are interacting with an AI system (Article 50(1)) and mark synthetic audio, image, video and text outputs in a machine-readable, detectable format (Article 50(2)). Article 50(1) 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". Under paragraph 70 of the guidelines, Article 50(2) needs both marking and detection. Recital 133 names watermarks, metadata identifications, cryptographic methods for proving provenance and authenticity, logging methods and fingerprints as techniques, and the transparency code asks signatories for at least two marking layers: digitally signed metadata and an imperceptible watermark.

Paragraph 4: What Deployers Disclose about Generated Content

Deployers disclose that deep fake image, audio or video content is artificially generated or manipulated, and disclose AI-generated text published to inform the public on matters of public interest (Article 50(4)). A deep fake, under Article 3(60), is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful, so text is never a deep fake. The duty shrinks to a disclosure that does not hamper the work when the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, and paragraph 122 of the guidelines adds that the informative character prevails when informative and creative content combine. The text duty lapses where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences, or where the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication.

Paragraph 3: Notices for Emotion Recognition and Biometric Categorisation

Deployers of an emotion recognition or biometric categorisation system inform the natural persons exposed to it of its operation and process their personal data under Regulations (EU) 2016/679 and 2018/1725 and Directive (EU) 2016/680 (Article 50(3)). Paragraph 104 of the guidelines applies the notice to any biometric categorisation, such as age or gender classification, whether high-risk or not, unless Article 5(1)(g) prohibits it. Paragraph 102 states that all emotion recognition systems are also high-risk unless Article 5(1)(f) prohibits them.

Article 5(1)(f) has prohibited emotion recognition in the areas of workplace and education institutions since 2 February 2025, except for medical or safety reasons, so emotion recognition at work and in education is a ban case, not a notice case. The remaining emotion recognition systems fall under Annex III, point 1(c), and carry the Chapter III Sections 1 to 3 duties from 2 December 2027 (Article 113, third paragraph, point (c)(i)) in addition to the notice.

Paragraphs 5 to 7: Manner of Information, Other Law and Codes after the Digital Omnibus

Every Article 50 notice is given in a clear and distinguishable manner at the latest at the first interaction or exposure and conforms to the applicable accessibility requirements (Article 50(5)). Paragraph 144 of the guidelines states that Article 50 imposes no distinct or additional accessibility requirements; those come from Directives (EU) 2016/2102 and 2019/882. The Digital Omnibus on AI replaced only paragraph 7 of Article 50 and left paragraphs 1 to 6 untouched.

Article 50(6) leaves Chapter III and other transparency obligations under Union or national law unaffected. Paragraph 7, as replaced, provides for codes of practice on the detection, marking and labelling of artificially generated or manipulated content: the Commission assesses their adequacy, taking utmost account of the opinion of the Board (Article 56(6) procedure), and may adopt an implementing act with common rules if a code is inadequate (Article 98(2)). Before 27 July 2026, paragraph 7 named the AI Office and provided for approval of codes by implementing act.

Recital 41 of Regulation (EU) 2026/1744 states that codes of practice "have limited legal effect, and in particular do not grant a presumption of conformity".

Which Article 50 Duty Applies? The Decision Logic in Five Questions

Five questions decide which Article 50 duty applies: interaction with natural persons, synthetic content, emotion recognition or biometric categorisation, deep fake output, and AI-generated text published on matters of public interest. The questions give an orientation, not a legal assessment, and each yes leads to one paragraph:

1. Interaction with natural persons? Yes: provider disclosure under Article 50(1).
2. Synthetic audio, image, video or text? Yes: provider marking and detection under Article 50(2).
3. Emotion recognition or biometric categorisation? Yes: deployer notice under Article 50(3).
4. Deep fake? Yes: deployer disclosure under Article 50(4), first subparagraph.
5. Text published to inform the public on matters of public interest? Yes: deployer disclosure under Article 50(4), second subparagraph, unless the law-enforcement exception or the human review exception applies.

Several paragraphs can apply to one system, and every yes also brings in Article 50(5). Illustration (our reading): a chatbot that also generates images can fall under questions 1 and 2, and its deployer under question 4 when a generated image is a deep fake. Paragraph 30 of the guidelines splits question 1 into four elements: an AI system, intended to interact, directly, with natural persons.

Article 50 Application Dates: 2 August 2026 and the 2 December 2026 Transition

Article 50 applies from 2 August 2026 under Article 113, second paragraph, and the Digital Omnibus on AI did not move that date. The Omnibus added a single transition, Article 111(4), inserted by Article 1, point (39)(b), of Regulation (EU) 2026/1744. Other provisions follow other application dates under Article 113, such as 2 February 2025 for Chapters I and II and 2 December 2027 for Chapter III Sections 1 to 3 for Annex III systems. The dated steps for Article 50 run from the 2025 consultation to the 2028 review:

Only Article 50(2) has a transition, lasting four months for generative systems already on the market (recital 38 of Regulation (EU) 2026/1744). Paragraph 153 of the guidelines adds that a partly interactive, partly generative system meets Article 50(1) from the general application date, without waiting for the marking deadline.

Article 112(2) obliges the Commission to evaluate the Article 50 list by 2 August 2028 and every four years. The yearly review of Article 112(1) covers Article 5 and Annex III, not Article 50.

What Article 50 Does Not Exempt: Open-Source Systems, Older Systems and Earlier Content

Three exclusions known from other parts of the act do not lift Article 50: the open-source exclusion of Article 2(12), the legacy rule of Article 111(2) and any general grace period for existing content. Article 2(12) excludes AI systems released under free and open-source licences unless they are placed on the market or put into service as high-risk AI systems or as an AI system that falls under Article 5 or 50, and paragraph 23 of the guidelines confirms this for Article 50. Illustration (our reading): an open-source image generator can fall under Article 50(2) like a proprietary one.

Article 111(2), the transitional rule for high-risk systems already on the market, does not apply to Article 50 obligations (guidelines, footnote 50). Article 111(4) covers marking and detection under paragraph 2 alone, with no grace period for paragraphs 1, 3, 4 or 5. Illustration (our reading): a chatbot placed on the market in 2025 can fall under Article 50(1) from the general application date.

Article 50 does not require retroactive labelling of content generated before 2 August 2026 (guidelines, paragraph 154). Text generated earlier but published on or after that date carries the Article 50(4) disclosure when it informs the public on matters of public interest, as the Commission Q&A also states.

The Article 50 Guidelines and the Transparency Code: Status on 24 October 2026

The Commission guidelines on Article 50 exist as a text whose content the Commission approved and published on 20 July 2026 by Communication C(2026) 5054 final; their formal adoption in all language versions is pending as of 24 October 2026. The text rests on Article 96(1)(d), is non-binding, and according to C(2026) 5054 final applies only once formally adopted in all language versions. The register of EU AI Act guidelines places it among all other Commission guidelines under the act.

The guidelines run to 155 paragraphs on 51 pages and followed a draft of 8 May 2026, open for consultation until 3 June 2026. Five official documents frame Article 50 as of 24 October 2026:

The final Code of Practice on Transparency of AI-generated Content covers Article 50(2), (4) and (5) in two sections of four commitments each: Section 1 for providers and Section 2 for deployers. The AI Office will consider facilitating formal updates at least every two years. The code is voluntary, and its text states that adherence "does not constitute conclusive evidence of compliance".

Organisations sign whole sections, and the Commission signatory list, last updated on 24 September 2026, shows 95 signatories of Section 1 and 192 of Section 2. Code task forces were established on 30 September 2026 with over 90 founding signatories, and use of the EU icons by a non-signatory does not indicate adherence. This transparency code is distinct from the General-Purpose AI Code of Practice of 10 July 2025, which serves Chapter V.

Article 50 Enforcement and Fines: Market Surveillance Authorities, the Commission and the EDPS

National market surveillance authorities enforce Article 50 from 2 August 2026, the AI Office acts for systems under its exclusive competence, and the EDPS supervises Union institutions. Article 70 required Member States to designate at least one market surveillance authority by 2 August 2025, and the national market surveillance authorities hold their enforcement powers from 2 August 2026, according to the Commission. In Germany, as of 24 October 2026, the Bundesnetzagentur is the default market surveillance authority, also for Article 50, under § 2(1) KI-MIG (BGBl. 2026 I Nr. 223 of 28 July 2026), and BaFin is competent for AI in regulated financial activity under § 2(3) KI-MIG.

The AI Office is exclusively competent under Article 75(1), as replaced by Regulation (EU) 2026/1744, for two groups of systems: (a) AI systems based on a general-purpose AI model where model and system come from the same provider or from providers in the same undertaking, except (i) Annex I products, (ii) Annex III point 2 systems, (iii) AI systems provided by law enforcement authorities, border management authorities and financial institutions, insofar as they fall under Article 74(6), and (iv) Annex III point 8 systems on the administration of justice; and (b) AI systems that constitute or are integrated into a very large online platform or very large online search engine designated under Regulation (EU) 2022/2065. Deployers fall under the AI Office only where they are also the provider or part of the same undertaking (Article 75(1), second subparagraph).

The European Data Protection Supervisor acts as market surveillance authority for Union institutions, bodies, offices and agencies under Article 50 (Commission Q&A; guidelines, paragraph 151), except for the Court of Justice acting in its judicial capacity (Article 74(9)). The maximum fines differ by addressee:

Commitments under an adequate code of practice can be a mitigating factor under Article 99(7), point (e) (guidelines, paragraph 149).

The EUR 15 million cap is the maximum that national penalty rules provide for (Article 99(1)); Chapter XII has applied since 2 August 2025 and reaches Article 50 breaches since Article 50 applies. The full system of AI Act fines also covers the Article 5 tier and Article 101. Any natural or legal person can complain to the relevant market surveillance authority under Article 85, and the AI Act Complaints Tool, which is not anonymous, covers only systems under AI Office competence.

How to Prepare for Article 50: Compliance Steps and Checklist for Providers and Deployers

Five steps prepare a provider or deployer for Article 50, which already applies: inventory the functions, assign the paragraph and the role, design the notices, implement marking or labelling, and document the measures. Each step maps to a paragraph:

1. Inventory every interactive, generative, emotion recognition, biometric categorisation or deep fake function.
2. Assign the role: provider for paragraphs 1 and 2, deployer for paragraphs 3 and 4.
3. Design each notice to meet Article 50(5).
4. Implement marking and detection as provider, or disclosure as deployer.
5. Document the measures, exceptions relied on and code commitments.

Providers of generative systems placed on the market before 2 August 2026 put marking first, because their Article 111(4) deadline is the only Article 50 date still ahead. A step-by-step programme for EU AI Act compliance places these steps inside the wider duties of the act. Every other Article 50 duty already applies.

Signing the transparency code is optional and gives no presumption of conformity. Commitment 4 in Section 2 of the code asks deployers for an editorial policy that identifies the person with editorial responsibility.

Records of notices, marks and exceptions relied on let an operator answer a market surveillance authority. The EU AI Act checklist sorts these duties by role and risk class.

Article 50 Transparency: Related Questions

Article 50 operates alongside the Chapter III high-risk rules, the Chapter V model duties, the GDPR and the Digital Services Act, the territorial scope of Article 2 and Article 4 AI literacy.

Do Article 50 Duties Apply on Top of the High-Risk Rules?

Article 50 duties apply on top of the high-risk rules, because Article 50(6) leaves Chapter III unaffected.

The Chapter III duties for high-risk AI systems and the Article 50 duties can attach to the same system. Article 13, "Transparency and provision of information to deployers", is a separate high-risk requirement and not part of Article 50.

How Does Article 50 Differ from the Chapter V Duties for General-Purpose Models?

Article 50 binds providers and deployers of AI systems, while Chapter V binds providers of general-purpose AI models.

The Article 53 duties of providers of general-purpose AI models include, under Article 53(1)(b), information and documentation, at minimum Annex XII, for providers integrating the model into an AI system. The General-Purpose AI Code of Practice serves Chapter V, not Article 50.

How Does Article 50 Interact with the GDPR and the Digital Services Act?

Article 50 adds to the GDPR and the Digital Services Act and replaces neither.

Article 50(3) requires processing under Regulation (EU) 2016/679, one of the points where the two regulations overlap in the comparison of AI Act vs GDPR. Recital 136 explains how the Article 50 duties relate to the Digital Services Act.

Do Non-EU Providers and Deployers Have to Meet Article 50?

Non-EU providers and deployers meet Article 50 whenever the act reaches them.

Article 2(1)(a) covers providers placing AI systems on the Union market irrespective of establishment, and the EU AI Act for non-EU companies also turns on Article 2(1)(c). That point covers third-country providers and deployers where the output of the system is used in the Union.

Is Article 4 AI Literacy Part of the Transparency Rules?

Article 4 AI literacy is a separate duty, not part of Article 50.

The duty of AI literacy under Article 4 applies since 2 February 2025, and since 27 July 2026 providers and deployers "shall take measures to support the development of AI literacy". Article 99(4) sets no EU fine tier for Article 4.

Does a Label Make AI-Generated Content Lawful?

A label does not make AI-generated content lawful: recital 137 states that compliance with Article 50 does not mean that the content or its use is lawful.

Article 5(1), points (ba) and (bb), on AI-generated intimate images and child sexual abuse material apply from 2 December 2026, whatever the label. Other transparency obligations under Union or national law also remain applicable under Article 50(6).

Frequently Asked Questions about Article 50 Transparency Obligations

What does Article 50 of the AI Act require?

Providers disclose AI interaction (paragraph 1) and mark synthetic outputs (paragraph 2); deployers inform persons exposed to emotion recognition or biometric categorisation (paragraph 3) and disclose deep fakes and public-interest text (paragraph 4).

When do the Article 50 transparency obligations apply?

From 2 August 2026, under Article 113, second paragraph. Providers of generative AI systems placed on the market before that date have until 2 December 2026 to meet the Article 50(2) marking duty (Article 111(4)).

Are the Article 50 guidelines binding?

No. The Commission approved and published their content on 20 July 2026 (C(2026) 5054 final), their formal adoption in all language versions is pending as of 24 October 2026, and they are non-binding.

Do open-source AI systems have to comply with Article 50?

Yes. The open-source exclusion of Article 2(12) does not apply to AI systems that fall under Article 50, so open-source providers and deployers meet the same transparency duties.

Is the EU icon for AI-generated content mandatory?

No. The EU icons are optional and free to use without attribution. The Article 50(4) disclosure itself is mandatory for deployers of deep fakes and of public-interest text, whatever label design they choose.

What is the fine for breaching Article 50?

Up to EUR 15 000 000 or 3 % of total worldwide annual turnover of the preceding financial year, whichever is higher (Article 99(4), point (g)); the lower amount for SMEs and small mid-caps; up to EUR 750 000 for Union institutions (Article 100(3)).