Article 50 transparency — deployer guide for 2 August 2026

The four sub-paragraphs

1. Art 50(1) — interaction notice. Providers must ensure AI systems intended to interact directly with natural persons are designed so that those persons are informed they are interacting with an AI system, unless this is obvious from the circumstances and context of use. Does not apply to AI systems authorised by law for criminal-offence purposes, subject to safeguards.
2. Art 50(2) — machine-readable marking of AI output. Providers of generative AI systems must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. Solutions must be effective, interoperable, robust and reliable as far as technically feasible.
3. Art 50(3) — emotion recognition / biometric categorisation notice. Deployers must inform natural persons exposed to the system of its operation, and process personal data in accordance with the GDPR and the Law Enforcement Directive. Exemptions narrowly cover legally authorised criminal-offence detection.
4. Art 50(4) — deep-fake and public-interest text disclosure. Deployers of AI generating deep fakes must disclose the content has been artificially generated or manipulated. For text published with the purpose of informing the public on matters of public interest, deployers must disclose that the text has been artificially generated or manipulated, with limited exceptions (e.g. human editorial review and responsibility).

Application date and transitional easing

Article 50 applies from 2 August 2026. The May 2026 AI Omnibus provisional agreement contemplates a transitional period through 2 December 2026 for generative AI systems already on the market before 2 August 2026 to comply with the machine-readable marking requirement under Article 50(2). Until that text is formally adopted and published in the Official Journal, the binding date in the Regulation remains 2 August 2026 for all sub-paragraphs.

On 8 May 2026 the European Commission published draft Guidelines on the implementation of Article 50 transparency obligations, clarifying overlap with the Digital Services Act and providing examples for chatbots, AI summaries and synthetic media.

Provider vs deployer split

• Provider — Art 50(1) design duty and Art 50(2) technical marking. The machine-readable signal (C2PA Content Credentials, SynthID, cryptographic hash, embedded metadata) is produced by the model.
• Deployer — Art 50(3) and 50(4) user-facing disclosure. Whether the chatbot banner is shown, whether the deep-fake watermark is preserved through your CMS, whether candidates and employees are told an AI assistant drafted their communication.

Deployer evidence checklist

• Inventory of every generative AI tool whose output reaches an external party.
• Vendor attestation that Art 50(2) marking is applied and that the marking survives common transformations (copy/paste, PDF export, screenshot).
• Standard disclosure text for chatbots and AI assistants — in the user’s language.
• SOP for AI-generated text on public-interest topics: human editorial review or visible disclosure.
• Deep-fake disclosure protocol for training videos and synthetic avatars.
• Emotion-recognition disclosure flow where such systems are used in any non-medical context.
• AI literacy training under Art 4 covering Art 50 obligations.
• Internal escalation path for complaints about undisclosed AI use.

Common misconceptions

• “Article 50 only applies to high-risk AI.” Wrong — Article 50 is a separate transparency regime that applies regardless of risk class.
• “The vendor handles it.” The vendor handles Art 50(2) technical marking. The deployer remains responsible for Art 50(1), (3), (4) disclosures and for documenting that vendor-side marking is operative.
• “The Digital Omnibus delayed Article 50.” No. The provisional agreement covers high-risk Annex III timing and a transitional easing for pre-existing generative systems under Art 50(2); it does not move the headline 2 August 2026 application date for Article 50.

Related EU guides

• EU AI Act for recruitment AI
• Deployer vs provider
• AI inventory requirements
• Fines under the EU AI Act

Sources

• Regulation (EU) 2024/1689, Articles 50, 113 — EUR-Lex: eur-lex.europa.eu/eli/reg/2024/1689/oj
• European Commission, Code of Practice on Transparency of AI-Generated Content (draft guidelines): digital-strategy.ec.europa.eu