EU AI Act Article 50
EU AI Act Article 50 generally applies from August 2, 2026 and assigns separate provider and deployer transparency duties. The adopted Digital Omnibus awaits Official Journal publication and entry into force. Once effective, Article 111(4) gives providers of generative AI systems placed on the market before that date until December 2, 2026 for Article 50(2) compliance.
In code
// The disclosure surface is your UI; the RECORD that it fired is audit-chain evidence.
// append() canonicalizes the payload, hash-chains it onto the tenant's tip, and mints a
// write-once WORM anchor in the same call — so the disclosure log can't be quietly edited.
await chainStore.append(accountId, {
event: "ai.disclosure.shown",
clause: "eu-ai-act/art-50-1",
surface: "support-chat",
sessionId,
disclosureVersion: "2026-07-10", // the versioned copy shown to the user
});
// Later — for the evidence bundle, or a regulator's question:
const result = await chainStore.verify(accountId);
// → { valid: true, brokenAt: null } — a non-null brokenAt surfaces tamper, insert, reorder,
// or truncation.How it holds
It applies to ordinary products, not just high-risk systems
Article 50(1) covers systems intended to interact directly with people; Article 50(2) separately covers providers of systems generating specified synthetic content. Each duty has conditions and exceptions. Article 50 generally applies from August 2, 2026. Under the adopted Digital Omnibus text, once effective, the targeted Article 50(2) transition for qualifying systems placed on the market before that date ends December 2, 2026.
Evidence supports the legal inquiry; it does not decide it
A WORM audit trail can preserve evidence that a disclosure event was recorded. It is supporting evidence, not a substitute for the disclosure, timing, accessibility, marking, and detection requirements, and it does not by itself establish that Article 50 was satisfied.
Marking and detection are one technical solution
Article 50(2) requires covered synthetic audio, image, video, and text outputs to be machine-readably marked and detectable, subject to express exceptions. The final guidance permits compliant marking post hoc, at the underlying model, or during inference.
Pre-existing content has a separate cutoff
Article 50(2) outputs and Article 50(4) deepfakes generated or manipulated before August 2, 2026 do not require retroactive marking or labelling. Public-interest text receives that treatment only when it was both generated or manipulated and published before August 2.
Frequently asked
Does Article 50 apply to my SaaS chatbot?
When does Article 50 become enforceable?
Do pre-existing outputs need retroactive labels?
Does Caisson make my product Article 50 compliant?
Primary sources
Sources verified 2026-07-27.
- Regulation (EU) 2024/1689 on EUR-Lex — Articles 3(3)–(4), 50(1)–(5), 99(4), 99(6), 100(3), and 113; OJ pp. 82–83, 115–117, and 123/144
- European Commission final-guidelines landing page — Page header “Publication 20 July 2026”; introductory paragraph beginning “The Commission adopted these guidelines”
- European Commission final Article 50 guidelines — Paragraphs (5), (6), (69)–(74), (151), and (153)–(154), pp. 3–4, 24–25, and 49–50
- European Commission quick facts on AI transparency — “Enforcement and penalties” → “Surveillance authorities”, “Penalties”, and “Exceptions”
- Council adopted Digital Omnibus legislative text — Article 111(4), p. 90/102, and Article 4, p. 101/102: systems “placed on the market before 2 August 2026”; compliance with Article 50(2) by 2 December 2026; entry into force on the third day after Official Journal publication
- Council final-approval and next-steps notice — Page header dated 29 June 2026; opening paragraph beginning “Today, the Council gave its final green light”; “Next steps” paragraph on Official Journal publication and entry into force