EU AI Act Article 50 transparency obligations

Article 50 sits in Chapter IV of the AI Act under the heading “Transparency obligations for providers and deployers of certain AI systems.” It is a directly applicable EU regulation provision, not soft guidance.

Article 113 of the Regulation states that the Act applies from 2 August 2026, subject to earlier and later dates for other chapters. Article 50 falls under that general application date. The Commission’s Article 50 FAQ repeats the same calendar fact: Article 50 applies as from 2 August 2026.

The Commission’s 31 July 2026 enforcement announcement states that from 2 August 2026 the AI Office, together with national authorities, begins enforcing the AI Act, and that the same date starts transparency rules requiring certain systems to tell users when they are interacting with AI and when content has been generated or altered by it.

Article 50 assigns different duties to providers and deployers. Providers design and place systems on the market. Deployers use systems under their authority, excluding purely personal non-professional use, as explained in the Commission FAQ.

Paragraph 1 requires providers to ensure that AI systems intended to interact directly with natural persons are designed so those persons are informed they are interacting with an AI system, unless that fact is obvious to a reasonably well-informed, observant and circumspect person in context. Law-enforcement authorised systems have a narrow carve-out.

Paragraph 2 requires providers of systems that generate synthetic audio, image, video or text, including general-purpose AI systems, to mark outputs in a machine-readable format so they are detectable as artificially generated or manipulated. Solutions must be effective, interoperable, robust and reliable as far as technically feasible. Assistive standard-editing functions and certain law-enforcement uses are excluded.

Paragraph 3 requires deployers of emotion-recognition or biometric-categorisation systems to inform exposed natural persons and to process personal data under the applicable Union data-protection instruments, again with a law-enforcement carve-out.

Paragraph 4 requires deployers to disclose deepfakes (AI-generated or manipulated image, audio or video content resembling existing persons, objects, places, entities or events that would falsely appear authentic). Artistic, creative, satirical or fictional works get a lighter disclosure that must not hamper enjoyment of the work. The same paragraph requires disclosure for AI-generated or manipulated text published to inform the public on matters of public interest, unless human review or editorial control plus editorial responsibility applies, or a law-enforcement carve-out applies.

Paragraph 5 requires the information in paragraphs 1 to 4 to be clear and distinguishable at the latest at first interaction or exposure, and to meet accessibility requirements. Paragraph 6 preserves Chapter III high-risk duties and other Union or national transparency laws. Paragraph 7 tasks the AI Office with encouraging codes of practice on detection and labelling, with Commission implementing powers if a code is inadequate.

• Provider · Art. 50(1): inform users of direct AI interaction unless obvious.
• Provider · Art. 50(2): machine-readable marks on covered synthetic outputs.
• Deployer · Art. 50(3): notice for emotion recognition / biometric categorisation.
• Deployer · Art. 50(4): disclose deepfakes; label covered public-interest AI text.
• All covered duties · Art. 50(5): clear notice by first exposure; accessibility.

From 2 August 2026, providers and deployers must comply with Article 50. The Commission FAQ states that a limited grace period applies only to AI systems placed on the market before 2 August 2026, and only for the marking and detection obligation in Article 50(2). Those providers must comply with that marking duty from 2 December 2026.

The FAQ also states that content generated prior to 2 August 2026 does not need to be labelled retroactively, while encouraging relevant deployers to label where possible.

The 31 July 2026 announcement adds that chatbots and other interactive AI systems must tell users they are dealing with AI; deepfakes must be labelled; and AI-generated or altered content must carry machine-readable marks. It also records that more than 180 organisations had signed the Code of Practice on transparency of AI-generated content that operationalises those rules.

The Commission FAQ states that compliance is mainly enforced by national competent market surveillance authorities. The AI Office’s role is limited: systems built on general-purpose AI models where the same entity provides system and model, or systems integrated into designated very large online platforms or search engines under the DSA. The European Data Protection Supervisor enforces for EU institutions, bodies and agencies.

The FAQ states that fines can reach up to 15 million euros or 3% of total worldwide turnover for the preceding financial year, with proportionality for SMEs and small mid-caps.

Article 50 is a transparency chapter. It does not itself contain the prohibited-practices list in Article 5, the high-risk conformity regime in Chapter III, or the general-purpose model documentation and systemic-risk duties in Chapter V.

Machine-readable marking under Article 50(2) is a provider technical duty. Visible deepfake disclosure under Article 50(4) is a deployer duty. The FAQ states deployers cannot rely only on the provider’s machine-readable mark to meet deepfake disclosure.

The Code of Practice on transparency of AI-generated content is a voluntary demonstration path for marking and labelling duties under Article 50(2), (4) and (5). The separate GPAI Code of Practice addresses model-level documentation under Articles 53 and 55. The FAQ treats them as complementary, not interchangeable.

Edge cases turn on Commission guidelines, codes of practice, national market-surveillance practice and later Omnibus or delegated acts. This page tracks the Regulation text and Commission FAQ language indexed in the evidence graph.

Whether a given product is “obviously” an AI interaction, qualifies as a deepfake, or benefits from editorial-control exemptions is a facts-and-guidelines question this page does not decide for named products.

The following actors are linked to this vehicle in the graph. Their presence here records institutional role on the page, not an independent finding that each actor has completed a compliance duty.

• European Commission AI Office (government): The Commission body responsible for implementing and enforcing the EU AI Act’s general-purpose AI rules and coordinating the wider regime.

Each block below is an explicit evidence row: a claim that points to one supporting passage. Read the quote and locator before treating the paraphrase as settled.

• Article 50 transparency duties entered into application. This establishes the operative date for EU duties covering disclosure of AI interaction and specified synthetic content. Operational note: It separates current legal obligations from future AI Act deadlines that were moved by later amendments.
• EU rules require detectable machine-readable marks. The duty goes beyond a visible notice. Providers must make covered outputs technically detectable as generated or altered. Operational note: Technical provenance is a distinct policy instrument from user-facing deepfake labels.

Dated events attached to this vehicle. Each entry is a graph event with its own update page when present.

• 2026-08-02 · enforcement · EU Article 50 transparency duties enter application. The development moved disclosure and technical-marking duties from implementation planning into the operative phase. Consequence recorded: Changed the immediate compliance position for covered providers and deployers.
• 2026-12-02 · deadline · Limited Article 50 legacy-system grace period ends. The deadline does not defer all Article 50 obligations. It applies narrowly to specified marking and detection duties for pre-existing systems. Consequence recorded: Closes the remaining transition identified in the Commission’s Article 50 guidance.

A vehicle is an instrument. Asks are the outcomes the instrument is treated as advancing in the AI Tribune graph.

• Require machine-readable marking of AI-generated or altered content Status: in-force. Require covered providers to attach detectable technical provenance markers to synthetic or materially altered content.
• Require visible disclosure for deepfakes and public-interest synthetic media Status: in-force. Require people encountering covered synthetic media to receive a clear, visible disclosure rather than relying only on hidden metadata.

These documents are the originals behind the evidence rows. Prefer the document text over secondary paraphrase when the two diverge.

• Transparency obligations under Article 50 of the AI Act (Official guidance, 2026-07-24). Commission guidance on scope, timing, enforcement and the limited legacy-system grace period.
• Commission starts enforcing AI Act rules and new transparency requirements (Official announcement, 2026-07-31). The Commission’s account of the enforcement phase beginning on 2 August 2026.

The following rows are labeled inference in the evidence graph. They combine more than one passage and must not be read as single-source facts.

• EU and Chinese rules converge on provider-side labelling: Both regimes impose provider-facing duties, although their legal scope and enforcement structure differ. Limit: This is cross-source analysis. It should not be read as a claim that terminology, exceptions or enforcement powers are identical.