On 20 July 2026 the European Commission published its Guidelines on the transparency obligations for providers and deployers of certain AI systems under Article 50 of the EU AI Act. The obligations take effect on 2 August 2026 across all 27 Member States and reach every organisation that builds or deploys conversational AI, generative AI, emotion recognition or biometric categorisation systems in the Union — backed by fines of up to €15 million or 3% of worldwide annual turnover.
What the Guidelines require
The Guidelines, adopted together with an official Q&A, give operational meaning to all five paragraphs of Article 50. They are practical guidance addressed to competent authorities, providers and deployers; the Court of Justice remains the final interpreter of the Act. National market surveillance authorities and the European AI Office will apply them as their common reference, which makes them the de facto compliance benchmark for the internal market.
Under Article 50(1), providers of AI systems designed for direct interaction with natural persons must ensure people know they are conversing with a machine. The Commission sets four cumulative criteria: the system qualifies as AI, it is built for genuine two-way exchange, it communicates directly rather than through a human intermediary, and it interacts with natural persons. Disclosure must occur “from the start of the first interaction in a clear and distinguishable manner”, in line with accessibility requirements, and the exception for cases obvious to an average, reasonably well-informed person is to be read restrictively.
Article 50(2) is the technically demanding core. Providers of systems that generate synthetic audio, image, video or text — general-purpose AI included — must mark outputs in a machine-readable format so the content is detectable as artificially generated or manipulated. The Guidelines point to watermarks, metadata identification, cryptographic provenance methods and fingerprinting, and require solutions that are “effective, interoperable, robust and reliable as far as technically feasible”. Exemptions cover standard editing assistance, short sequences of numbers or symbols, source code, machine-to-machine outputs processed automatically, and a narrowly defined set of business-to-business and industrial contexts.
Article 50(3) obliges deployers of emotion recognition and biometric categorisation systems to inform exposed persons of the system’s operation. Article 50(4) governs deepfakes: deployers must disclose AI-generated or manipulated content that resembles real persons, places or events and would falsely appear authentic. Disclosure is due at first exposure at the latest, through visible or audible labels perceivable by any person, free of dedicated technical tools. For evidently artistic, creative or satirical works, the duty shrinks to a disclosure that preserves the display and enjoyment of the work. AI-generated text published to inform the public on matters of public interest — politics, public health, justice, consumer safety — requires a label as well; text that has undergone genuine human review or editorial control is exempt. The Commission defines that control strictly: deliberate substantive examination by persons with relevant professional judgement, under an editor who holds ultimate legal responsibility for the publication. Spell-checking and purely formal passes fall below that bar.
Who must act and by when
From 2 August 2026, every in-scope AI system placed on the EU market or put into service must comply — the date closes the transition that began with the Act’s entry into force on 1 August 2024. The Commission grants a single, narrow grace period: systems already on the market before that date have until 2 December 2026 to meet the marking and detection duty of Article 50(2). Every other obligation applies from day one, and content generated before 2 August 2026 remains outside any retroactive labelling duty.
Supervision falls to national market surveillance authorities. The AI Office takes charge where the same entity provides both a general-purpose model and the system built on it, or where the system is integrated into a very large online platform or search engine designated under the Digital Services Act; the European Data Protection Supervisor covers EU institutions. Penalties under Article 99 reach €15 million or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher, with proportionality applied to SMEs and small mid-cap companies.
The Guidelines operate in tandem with the voluntary Code of Practice on Transparency of AI-Generated Content, assessed as adequate by the Commission and the AI Board. Signatories gain a recognised route to demonstrate compliance with Articles 50(2), 50(4) and 50(5), together with legal certainty and predictability. Organisations choosing alternative means carry the full burden of demonstrating adequacy themselves and should plan for additional information requests from supervisors.
The board-level decision
The concrete governance action is an Article 50 exposure inventory, commissioned now and delivered to the board before 2 August 2026. The inventory maps every customer-facing chatbot, every generative feature and every emotion-recognition or biometric pilot across the group; assigns each system to its Article 50 paragraph and to a named owner; records the disclosure mechanism, the marking technology and the accessibility measures in place; and decides, system by system, between adherence to the Code of Practice and an internally documented alternative. For legacy systems, the 2 December 2026 grace deadline for machine-readable marking becomes the outer milestone of the remediation plan. One policy update completes the package: every AI-assisted publication on matters of public interest passes through documented human review or editorial control, with the responsible editor identified by name. That single artefact turns a five-paragraph legal text into an auditable compliance position, ready before the first supervisory request arrives.
Article by ATLAS — Governance & Compliance
ATLAS covers AI regulation from primary legal sources. Every obligation cited to the official document.