Code of Practice on Transparency of AI-generated Content (2026)¶
Reference
Citation: European Commission, European AI Office. "Code of Practice on Transparency of AI-generated Content" (10 June 2026). Type: guidance. Link: digital-strategy.ec.europa.eu.
What it is¶
The voluntary code provided for by Article 50(7) of the AI Act, through which signatories can demonstrate compliance with the marking duty in Article 50(2) and the labelling duty in Article 50(4). Section 1 commits providers of generative systems to two layers of machine-readable marking: digitally signed and time-stamped metadata where the format carries metadata, and an imperceptible watermark, required for free-form text above 200 tokens. Section 2 commits deployers to label deepfakes and published public-interest text, and to name a person or function holding editorial responsibility with published contact details. Adherence is voluntary and, in the code's own words, "does not constitute conclusive evidence of compliance".
Role in the record¶
- Grounds BP07: a concrete specification of machine-readable marking of AI-generated content, as signed and time-stamped metadata plus a watermark that is hard to separate from the content (Section 1, Sub-measures 1.1.1 and 1.1.2).
- Grounds BP04: a named person, role, and contact details for editorial responsibility, published where not already public (Section 2, Commitment 4).
- Qualifies BP09: the code requires a policy for human review but states that this "does not entail having to document individual instances of human review or editorial control over individual text publications" (Section 2, Commitment 4(b)).
Atom-level for/against detail and quotes are in the provenance data (assets/provenance.yml), keyed by practice atom.