AI Act: Code of Practice signing — D‑41 before your AI labels

D‑41 before AI labelling takes effect. On 22 June 2026, the Commission’s AI Alliance held an info and signing session for the Code of Practice on labelling AI‑generated content. Transparency obligations under Article 50 of the AI Act apply from 2 August 2026.

Key facts

On 22 June 2026 (14:00–16:00 CEST), the European Commission (AI Alliance/Futurium) briefed companies and organisations on the “Code of Practice on Transparency of AI‑generated content” and how to sign it. The goal: standardise the labelling of AI outputs (text, audio, image, video) ahead of the transparency go‑live on 2 August 2026.

This milestone directly affects deployers and providers in Luxembourg and the Greater Region (BE, FR, DE) who publish or distribute AI‑generated or AI‑manipulated content. From 2 August 2026, the AI Act’s general applicability starts; administrative fines can reach €35m or 7% of global turnover for certain infringements, as provided by the Regulation.

Legal framework and basis

The basis is Article 50 of the AI Act (Regulation (EU) 2024/1689), which requires: (1) clear information whenever a person interacts with an AI system (chatbots); (2) machine‑readable labelling and detectability of artificially generated or manipulated content (including by general‑purpose AI systems); (3) specific disclosure duties for “deepfakes”, with a lighter regime for manifestly artistic/satirical works subject to safeguards for third parties. For an overview, see the AI Act and its Article 50 framework.

What this changes for Luxembourg businesses

From 2 August 2026, any entity in Luxembourg (banking, asset management, industry, retail, media, public sector) deploying: (a) chatbots/AI assistants; (b) generative tools producing external content (posts, campaigns, customer support, images/videos, voice); or (c) AI post‑production solutions (upscaling, voice cloning, substantial editing) must demonstrate: clear user information; machine‑readable labelling (watermarking/metadata or content authentication); and an internal process for “deepfakes”.

The Code of Practice provides a fast, operational path to standardise labelling across complex publishing chains (agencies, CMS, DAM, social). Leaders in Luxembourg and neighbouring markets should launch technical and documentation work now. To structure and drive this, our AI governance and compliance service can support labelling design, disclosures and compliance records.

Concrete actions to take this week

• Map external AI uses: inventory every touchpoint where end‑users may see or hear AI‑generated/manipulated content (site, app, chatbot, callbot, emails, social media, marketing collateral, customer support) and qualify “light assistance” vs “substantial generation” to align with Article 50.
• Select and test labelling: implement a watermarking and/or metadata scheme (C2PA/manifests, XMP, ID3, subtitles) compatible with your CMS/DAM and platforms (YouTube, LinkedIn, X, Instagram). Document technical limits and define a fallback (visible disclosure) if machine‑readable labels are stripped downstream.
• Update notices & UX: add an explicit “you are interacting with an AI system” notice to interfaces; for content, insert a standardised disclosure (tag, banner, caption) and update editorial policies and publishing templates.
• Qualify deepfakes: set up a specific assessment and disclosure workflow (transformation log, preservation of originals, legal/marketing validation) and a lighter regime for manifestly artistic works while safeguarding third‑party rights.
• Governance and evidence: appoint an “AI transparency” owner (Marketing/Comms + DPO + Security), maintain a compliance register (technical choices, label robustness tests, scope of exemptions), and plan a response if labels are removed (takedown, correction, corrective comms). For local roll‑out specifics, see AI compliance in Luxembourg.
• Join the Code of Practice: enrol as a signatory via the AI Alliance/Futurium and schedule an internal audit at D+30 to verify effectiveness before 2 August 2026.

Next step

Rapid scoping of Article 50 duties, C2PA/metadata labelling and channel disclosures secures readiness before the deadline. Need an operational checkpoint? Tell us about your use cases and timelines.

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