EU AI Act Article 50 transparency

Article 50 of the EU AI Act (Regulation 2024/1689) imposes transparency obligations on providers and deployers of certain AI systems. The obligations become enforceable on 2 August 2026.

The European Commission's Code of Practice on Transparency of AI-Generated Content (first draft 17 December 2025; final ~June 2026) sets out what compliant implementations look like. Provenance ships disclosure templates aligned with the draft and provides machine-readable provenance metadata that survives re-encoding and distribution.

The four obligation tracks

Article 50 sets requirements across four distinct tracks. Each maps to a different set of AI systems and a different combination of visible disclosure + machine-readable mark.

Article 50(1)
AI interaction notice
Applies to: Providers of AI systems intended to interact directly with natural persons (chatbots, voice assistants, AI agents).
Visible requirement: Inform end users they are interacting with an AI system, unless this is obvious to a reasonably well-informed person from the circumstances and context of use.
Machine-readable requirement: Surface the AI system identifier in the response envelope (HTTP header, JSON-LD `dct:type`, or system message metadata) so accessibility tools and downstream agents can detect the AI source.
Examples: In-app chat assistant that helps users navigate the product · Voice agent that handles inbound support calls · Email triage bot that responds without human review
Article 50(2)
AI-generated content marking
Applies to: Providers of AI systems generating or manipulating text, image, audio, or video content (synthetic media).
Visible requirement: Mark output as artificially generated or manipulated in a way the end user can perceive — visible label, audio cue, or watermark depending on modality.
Machine-readable requirement: Apply machine-readable provenance metadata (C2PA-compatible signing, EXIF tags, ID3 frames, or robust watermarking) so detection mechanisms and downstream platforms can identify the content as AI-generated.
Examples: AI-generated marketing copy or product descriptions · AI-generated product photography for e-commerce · AI-generated audio for podcast production
Article 50(4) §1
Deep-fake disclosure
Applies to: Deployers of AI systems generating or manipulating content constituting a deep fake (image, audio, or video resembling real persons, objects, places, entities, or events).
Visible requirement: Disclose that the content has been artificially generated or manipulated in a clear and distinguishable way at the latest at the time of first interaction or exposure.
Machine-readable requirement: Combine the deep-fake disclosure with provenance signals — content credentials, signing chains, or watermarks — that survive re-encoding and distribution. The multilayer approach combines visible disclosure with invisible or machine-readable techniques.
Examples: AI-generated likeness in an ad creative · Voice cloning for customer support · Synthetic talking-head explainer videos
Article 50(4) §2
Public-interest content disclosure
Applies to: Deployers of AI systems generating or manipulating text published with the purpose of informing the public on matters of public interest.
Visible requirement: Disclose that the text has been artificially generated or manipulated, except where the AI-generated content has undergone human review or editorial control and where a natural person or legal person holds editorial responsibility for the publication.
Machine-readable requirement: Annotate the publication with machine-readable provenance + editorial-responsibility metadata (e.g. `prov:wasGeneratedBy` + `prov:wasAttributedTo`) so search engines, fact-checkers, and aggregators can surface the disclosure alongside the content.
Examples: AI-assisted news articles published by a media organization · AI-generated political analysis or commentary · AI-summarized legal or civic information

How Provenance helps

Each disclosure publishes inside your trust page with two layers:

Visible disclosure — pre-filled Markdown template per track that you edit and publish. Available in English, German, French, and Italian; see the templates.
Machine-readable mark — a JSON-LD payload at /trust/[slug]/export/jsonld with content hash, timestamp, and provenance metadata. Compatible with C2PA-style content credentials and standard search-engine crawlers.

Informational only — not legal advice. Article 50 obligations vary by AI system type and deployment role; consult counsel for your specific obligations. Last updated against the December 2025 Code of Practice draft.

EU AI Act Article 50 transparency

Article 50 of the EU AI Act (Regulation 2024/1689) imposes transparency obligations on providers and deployers of certain AI systems. The obligations become enforceable on 2 August 2026.

The four obligation tracks

Article 50 sets requirements across four distinct tracks. Each maps to a different set of AI systems and a different combination of visible disclosure + machine-readable mark.

Article 50(1)

AI interaction notice

Applies to: Providers of AI systems intended to interact directly with natural persons (chatbots, voice assistants, AI agents).

Visible requirement: Inform end users they are interacting with an AI system, unless this is obvious to a reasonably well-informed person from the circumstances and context of use.

Machine-readable requirement: Surface the AI system identifier in the response envelope (HTTP header, JSON-LD `dct:type`, or system message metadata) so accessibility tools and downstream agents can detect the AI source.

Examples: In-app chat assistant that helps users navigate the product · Voice agent that handles inbound support calls · Email triage bot that responds without human review

Article 50(2)

AI-generated content marking

Applies to: Providers of AI systems generating or manipulating text, image, audio, or video content (synthetic media).

Visible requirement: Mark output as artificially generated or manipulated in a way the end user can perceive — visible label, audio cue, or watermark depending on modality.

Machine-readable requirement: Apply machine-readable provenance metadata (C2PA-compatible signing, EXIF tags, ID3 frames, or robust watermarking) so detection mechanisms and downstream platforms can identify the content as AI-generated.

Examples: AI-generated marketing copy or product descriptions · AI-generated product photography for e-commerce · AI-generated audio for podcast production

Article 50(4) §1

Deep-fake disclosure

Applies to: Deployers of AI systems generating or manipulating content constituting a deep fake (image, audio, or video resembling real persons, objects, places, entities, or events).

Visible requirement: Disclose that the content has been artificially generated or manipulated in a clear and distinguishable way at the latest at the time of first interaction or exposure.

Machine-readable requirement: Combine the deep-fake disclosure with provenance signals — content credentials, signing chains, or watermarks — that survive re-encoding and distribution. The multilayer approach combines visible disclosure with invisible or machine-readable techniques.

Examples: AI-generated likeness in an ad creative · Voice cloning for customer support · Synthetic talking-head explainer videos

Article 50(4) §2

Public-interest content disclosure

Applies to: Deployers of AI systems generating or manipulating text published with the purpose of informing the public on matters of public interest.

Visible requirement: Disclose that the text has been artificially generated or manipulated, except where the AI-generated content has undergone human review or editorial control and where a natural person or legal person holds editorial responsibility for the publication.

Machine-readable requirement: Annotate the publication with machine-readable provenance + editorial-responsibility metadata (e.g. `prov:wasGeneratedBy` + `prov:wasAttributedTo`) so search engines, fact-checkers, and aggregators can surface the disclosure alongside the content.

Examples: AI-assisted news articles published by a media organization · AI-generated political analysis or commentary · AI-summarized legal or civic information

How Provenance helps

Each disclosure publishes inside your trust page with two layers:

Visible disclosure — pre-filled Markdown template per track that you edit and publish. Available in English, German, French, and Italian; see the templates.

Machine-readable mark — a JSON-LD payload at /trust/[slug]/export/jsonld with content hash, timestamp, and provenance metadata. Compatible with C2PA-style content credentials and standard search-engine crawlers.