AI Act D-31: transparency on 2 August, machine-readable marking by 2 December

Key point — On 29 June 2026, the EU Council approved the “Digital Omnibus”, confirming the start date of the AI Act’s transparency duties on 2 August 2026 (Article 50) and setting a grace period until 2 December 2026 for machine‑readable marking of AI content already in service before 2 August.

What happened

The “Digital Omnibus” fine‑tunes timelines but locks in the application of Article 50 across the EU from 2 August 2026. The grace period to deploy transparency solutions (marking/labelling) is shortened from 6 to 3 months for providers of artificially generated content, setting an additional milestone on 2 December 2026 for machine‑readable marking. On 10 June 2026, the European Commission published a Code of Practice to support implementation.

Legal basis

• AI Act — Article 50: inform people when they interact with an AI system (Art. 50(1)) and label generated or manipulated content, including deepfakes. For machine‑readable marking, an additional period runs until 2 December 2026 for systems already on the market before 2 August.
• Code of Practice (10/06/2026): a voluntary instrument with concrete measures (secure metadata, watermarks, editorial‑exception policies) to demonstrate compliance with Article 50.

What this means for Luxembourg businesses

• Who is in scope: entities operating in Luxembourg or targeting EU users that deploy chatbots/voicebots/assistants, or publish/integrate AI‑generated or manipulated content (text, images, audio/video), including marketing and non‑editorially reviewed public‑interest content.
• Risk: non‑compliance as of 2 August 2026 (no user notice, no clear AI/deepfake labelling). For machine‑readable marking, a practical milestone on 2 December 2026 for systems already in service; exposure to orders, content takedowns and national sanctions under the AI Act.
• Timeline: D‑31 to 2 August 2026 to activate “You are interacting with AI” notices and human‑readable labels; D‑154 to 2 December 2026 to roll out machine‑readable marking.

To structure the workstreams, consider dedicated support for AI Act compliance implementation to industrialise badges, notices and metadata across your journeys and content.

Immediate actions to take this week

• Map AI touchpoints: list chat/voicebots, widgets, IVR scripts and content generation; distinguish human‑facing interactions from back‑office automation; prioritise customer/employee‑visible channels.
• Enable immediate “AI in interaction” notice: by 2 August, add a clear notice in the user’s language at conversation start and in the UI (persistent badge + tooltip); document location, wording and deployment evidence.
• Label human‑readable AI content: state “AI‑generated content”/“AI‑modified”. For public‑interest deepfakes not editorially reviewed, add reinforced alerts (audio/video pre‑roll, visible watermark).
• Prepare machine‑readable marking (by 2 December): choose a CMS/CDN‑ and social‑compatible strategy (C2PA metadata, XMP, JSON‑LD schemas, imperceptible watermarking); run a pilot and test robustness after transcoding.
• Handle editorial exceptions: set an escalation path to an editorial lead for public‑interest content reviewed under editorial responsibility, and retain proof of review.
• Update privacy/compliance: adjust GDPR notices and train front‑office/marketing on the “AI interaction” announcement and deepfake reporting.
• Qualify vendors: require an Article 50 plan (badges, help pages, metadata, watermarks) and coverage metrics.
• Steer with KPIs and evidence: track % of interactions showing the AI notice, % of human‑readable labels, and by 2 December, % with machine‑readable marking verified by a third‑party tool; archive screenshots and C2PA manifests.

Review the detailed regulatory context on the AI Act transparency obligations, then plan your deployments. For hands‑on assistance, tell us about your timeline.

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