AI Act: 52 days to go before transparency duty (Article 50)

Summary — From 2 August 2026, Article 50 of Regulation (EU) 2024/1689 imposes transparency duties: clear user notices when interacting with AI, machine‑readable labels on generated or manipulated content, and deepfake disclosures. Political adjustments do not change this date, except for machine‑readable retro‑labeling of existing content, expected by 2 December 2026.

The facts

On 2 August 2026, Article 50 transparency requirements apply across the EU (including Luxembourg). The Commission issued draft guidelines on 8 May 2026, and Council/Parliament confirmed the obligations remain due on that date; only retro‑labeling of content already on the market is expected to move to 2 December 2026.

Legal basis

• AI Act Article 50: user information for AI interactions, labeling of generated/manipulated content, and special cases (emotions/biometrics, deepfakes) with context‑appropriate information. See our overview of the AI Act and Article 50.
• “Digital Omnibus” political deal (7 May 2026): some high‑risk items deferred to 2/12/2027, without affecting Article 50; retro‑labeling of existing content expected by 2/12/2026.
• Commission guidelines (8 May 2026): details on the AI notice, machine‑readable labeling, and limited exceptions (e.g., public interest).

What changes for Luxembourg companies

• Who is in scope: operators of chatbots, text/visual/audio/video generation, publishers of synthetic media, and users of emotions/biometrics techniques not classified as high‑risk. Deployers of third‑party solutions have their own information duty.
• Risk: non‑compliance with a directly applicable regulation, plus privacy risks. For GDPR alignment, see our GDPR resource.
• Timeline: 2 August 2026 for notices and labeling of new content; 2 December 2026 expected for retro‑labeling of legacy items.

Immediate action plan

• Map all AI touchpoints (chatbots, writing assistants, image/video generation, speech synthesis, filters/retouching, marketing posts) and separate new vs legacy content.
• Draft and test visible “You are interacting with AI” notices with a short explainer (purpose, limits, human contact point, safety guidance, escalation).
• Implement machine‑readable labeling (e.g., C2PA/Content Credentials), including pipeline logging and preparations for retro‑labeling.
• Deployer governance: contractual clauses with AI vendors to ensure labeling signals and API/SDK compatibility; a plan B if guidance is not followed.
• GDPR documentation: information notices, DPIA as needed, records, rights handling; align AI notices and legal basis. If you need support, explore our AI governance and compliance service.
• Supervision: first‑line controls, compliance KPIs (detectable labeling rate, time‑to‑fix), and an incident playbook.

Next steps

Legal, product, and communications teams should converge by end‑June to publish a short “AI & transparency” policy and enable notices/labels in production before 2 August. For a quick discussion with our experts, contact the Luxgap team.

Article generated by Luxgap regulatory watch. For tailored guidance on this topic, contact us.