AI Act Article 50: what you must disclose about AI-generated content, and when
Article 50 of the EU AI Act creates transparency obligations that become operative in 2026. For organizations that generate, publish, or distribute AI-assisted content, this is no longer a future compliance consideration — it is a present design and governance task. This brief maps what Article 50 requires, who is in scope, and what boards should be able to verify their organizations have built.
This is a published intelligence brief, not legal advice. Claims are tied to named instruments, with sources and a snapshot date below.
What Article 50 actually requires
Article 50 of Regulation (EU) 2024/1689 (the EU AI Act) imposes transparency obligations in four categories. The most directly relevant for organizations that publish or distribute AI-generated content are:
Providers of AI systems that generate synthetic content (Article 50(2)) must ensure the outputs are marked as artificially generated in a machine-readable format. This duty sits with the provider — the party that develops the generative AI system and places it on the market — which is a distinct role from a deployer that uses such a system. Providers and deployers carry different Article 50 duties, and the first governance step is knowing which role you occupy for each tool.
Deployers face two distinct Article 50(4) disclosure triggers — and the public-interest qualifier attaches to only one of them. First, deployers of AI systems that generate or manipulate deepfakes — image, audio, or video — must disclose that the content is artificially generated or manipulated; this trigger is not limited to matters of public interest (it carries limited carve-outs for artistic, creative, or satirical work). Second, deployers who publish AI-generated text to inform the public on matters of public interest must disclose that it is AI-generated — unless the text "has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility." That editorial-review exception for public-interest text is the legal hinge for the distinction below.
Snapshot (EU AI Act): Article 50 transparency duties become applicable on 2 August 2026 (Article 113). Instrument: Regulation (EU) 2024/1689, published in the Official Journal 12 July 2024. Marking of generative-AI content already on the market: grace period to 2 December 2026 under the provisional AI Omnibus agreement (7 May 2026).
The distinction that trips organizations up
“Fully AI-generated” and “AI-assisted” are treated differently under Article 50, but this distinction is not a free pass. It is a classification that must be defensible.
Fully AI-generated content is output produced by a generative AI system with no meaningful human editorial contribution — a text, image, or video where the AI system is the originating author and a human has not substantively revised or validated the content as their own.
AI-assisted content is content where a human author used AI tools to draft, research, or produce elements, but exercised substantive editorial judgment over the final output — checking sources, revising material, and taking professional responsibility for the claims made.
The governance implication: to classify content as “AI-assisted” rather than “AI-generated,” an organization must be able to show how the human contribution was exercised. A workflow where AI generates the draft and a human approves it without substantive revision is unlikely to qualify as “assisted” in a regulatory audit.
Who is in scope
The AI Act's territorial scope (Article 2) reaches providers that place AI systems on the EU market and deployers established or located in the EU — and, under Article 2(1)(c), providers and deployers in a third country where the output of the AI system is used in the Union. In practice an organization outside the EU can fall within scope where its AI-system output is used in the EU; the seat of incorporation is not the sole test. Whether a given activity is in scope is a legal question on the facts, not a general rule.
Content types covered include text, images, audio, and video. The emphasis in recitals and in the code of practice process is on content concerning elections, public health, economic conditions, and social cohesion — but the operative Article 50 text is broader than those examples.
What disclosure infrastructure looks like
A disclosure posture under Article 50 is not a policy memo. It is operational infrastructure with three components:
Content inventory: A current map of which content workflows in the organization use generative AI, what role AI plays in each, and how human editorial input is documented. This inventory must be maintained as workflows change.
Marking and labeling rules: A defined standard for machine-readable marking of AI-generated outputs — aligned with the harmonized label standard being developed under Article 50(7) — plus two human-readable disclosure protocols that match the Article 50(4) triggers: a deepfake protocol that labels AI-generated or manipulated image, audio, and video, and a public-interest text protocol that discloses AI-generated text (or records the human editorial review that exempts it).
Provenance record: A log or metadata system that allows the organization to demonstrate, for any piece of content, which workflow produced it and what the nature of human oversight was. This is the audit trail regulators will ask to see.
A disclosure posture is infrastructure, not a policy memo. The audit question is not “do you have a policy?” but “can you demonstrate, for this content, what workflow produced it and what human oversight looked like?”
The timeline, and what just changed
Article 50 becomes applicable on 2 August 2026. Two developments in the first half of 2026 sharpen the picture:
The Code of Practice on Transparency of AI-Generated Content — the voluntary instrument that operationalizes the Article 50(2) machine-readable marking duty — reached its final version on 10 June 2026 and is undergoing an adequacy assessment by the Commission and the AI Board. Once assessed as adequate, adherence to the Code is expected to give signatories a route to demonstrate compliance, while organizations that do not sign would need to show the adequacy of their own measures to market-surveillance authorities. The Commission has separately consulted on draft guidelines interpreting the Article 50 obligations.
The AI ("Digital") Omnibus reached a provisional political agreement on 7 May 2026. The general transparency duties under Article 50 are not postponed — they still apply from 2 August 2026 — but generative AI systems already placed on the market before that date are given until 2 December 2026 to meet the Article 50(2) machine-readable marking requirement. The Omnibus changes remain provisional pending formal adoption.
How Quentir reads it
Quentir’s Intelligence Library tracks the Article 50 operative dates, the Transparency Code of Practice and the Omnibus marking grace period, and the assisted-versus-generated taxonomy — each entry tied to a named instrument with a snapshot date. It is reading and readiness material for boards and counsel; it is not legal advice on your organization’s specific situation.
Sources: Regulation (EU) 2024/1689 Art. 50 & Art. 113 (OJ 12 July 2024); Code of Practice on Transparency of AI-Generated Content (final version 10 June 2026, under Commission/AI Board adequacy assessment); EU AI ("Digital") Omnibus provisional agreement (7 May 2026); European Commission draft guidelines on Article 50 transparency. Snapshot date: June 2026.
Stay current on AI regulation
Join the Intelligence Library waitlist for the Regulatory Monitor — what changed, what it means, dated on every entry.
This brief is published intelligence produced by Quentir Systems LLC. It does not constitute legal advice and creates no advisory or client relationship. Consult qualified legal and regulatory counsel for advice specific to your organization’s situation and jurisdiction.