Home / The Act / Chapter IV / Article 50

Article 50 — Transparency obligations for providers and deployers of certain AI systems

APPLIES AUG 2, 2026 — 12 DAYS 50(2) LEGACY SYSTEMS: DEC 2, 2026 AMENDED BY OMNIBUS
Chapter IV · Limited-risk tier
Source: EUR-Lex · CELEX 32024R1689
Verified 2026-07-16
PLAIN-LANGUAGE SUMMARY — ORIGINAL INTERPRETATION, NOT THE OFFICIAL TEXT

Article 50 is the transparency layer of the Act. If people interact with your AI, they must be able to tell. It carries four distinct duties — split below, because each lands on a different actor. These duties were not deferred by the Omnibus: they apply from 2 August 2026, with one narrow grace period for marking output of systems already on the market (2 December 2026).

50(1) Tell people they are interacting with AI Systems intended to interact directly with people must be designed so users know it’s AI — unless it’s obvious to a reasonably informed person. PROVIDERS · OBL-0051 AUG 2, 2026
50(2) Mark AI-generated output machine-readably Generative systems must mark synthetic audio, image, video and text so machines can detect it — watermarks, metadata, provenance. New systems from Aug 2; legacy systems get to Dec 2. PROVIDERS · OBL-0050 AUG 2 / DEC 2
50(3) Disclose emotion recognition & biometric categorisation People exposed to these systems must be informed, and personal data handled per GDPR. Narrow law-enforcement exceptions. DEPLOYERS · OBL-0052 AUG 2, 2026
50(4) Label deepfakes and AI-written public-interest text Deployers must disclose artificially generated or manipulated image, audio or video content, and AI-generated text published to inform the public — with artistic/editorial carve-outs. DEPLOYERS · OBL-0053 AUG 2, 2026
OFFICIAL TEXT Read verbatim on EUR-Lex ↗
The authoritative wording lives on EUR-Lex and is quoted there in all 24 languages. We deep-link to the paragraph level rather than reproducing the full text — so what you read here is always interpretation, and what you cite is always the law. The consolidated post-Omnibus version will be linked the day its CELEX publishes.
WHAT THE OMNIBUS CHANGED HERE
50(2) marking, legacy systems: systems on the market before 2 Aug 2026 get until 2 Dec 2026 to mark output (transition cut from 6 to 3 months). was: 2 Aug 2026
Everything else: 50(1) disclosure, 50(3) emotion-recognition/biometric disclosure, 50(4) deepfake labelling — unchanged, apply 2 Aug 2026.
Source: Omnibus final text (OJ pending, expected ≤ Jul 30) · verified 2026-07-16
HOW TO COMPLY

The Code of Practice on Transparency of AI-Generated Content (final, 10 Jun 2026) is the Commission-endorsed route to Art. 50 compliance for marking and labelling.

View in guidance registry →
RELATED
Obligations OBL-0050–0053 →GPAI hub (Art. 53 interplay) →Transparency tier in the risk model →"Deepfake" — Art. 3(60) →
PENALTY EXPOSURE
€15M or 3%
of worldwide turnover — Art. 99(4)(g). Enforceable from Aug 2, 2026.
← PREVIOUS · ART. 49 Registration NEXT · ART. 51 → Classification of general-purpose AI models as general-purpose AI models with systemic risk