EU AI transparency code sets marking and labelling playbook

The EU AI Act’s transparency regime under Article 50 applies from 2 August 2026, with the final Code of Practice giving companies a voluntary compliance framework for AI-generated content. The code operationalises the provider duty to mark AI-generated or manipulated output and the deployer duty to label deep fakes and AI-written public-interest text.

For providers, the code favours a multi-layered approach because no single marking technique is treated as sufficient in most online contexts. Signatories are expected to use at least two machine-readable layers, typically digitally signed metadata and imperceptible watermarking, while free-form text longer than 200 tokens must carry an imperceptible watermark. Providers must also preserve markings, prohibit circumvention, offer detection free of charge, and implement watermark-detection interoperability by 2 February 2027.

For deployers, the framework centres on a common EU icon using the capitalised acronym “AI” in English, with optional “GENERATED” or “MODIFIED” wording. Labels must be recognisable at first exposure and accessible, while audio-only content requires an audible disclaimer at the start. Existing systems placed on the market before 2 August 2026 have until 2 December 2026 to meet Article 50(2) marking and detection duties.