Skip to content

EU — Content Labeling and Provenance

Regulatory tiers

Section titled “Regulatory tiers”
  • Regulation (EU) 2024/1689 (AI Act): in force Aug 1, 2024; article 50 sets out the transparency and labeling obligations.
  • Core timeline for art. 50: applies from Aug 2, 2026 (general deployer obligations), consistent with the regulation’s staggered applicability.
  • Implementation code: the Commission published the final Code of Practice on Transparency of AI-Generated Content on 10 June 2026. It is voluntary and supports article 50 compliance from 2 August 2026.
  • Accompanying standards: the European standardisation organisations (CEN-CENELEC JTC 21) continue developing harmonised standards for “machine-readable and robust labeling”; once published, they will produce a presumption of conformity.
  • Adjacent regulations: the Digital Services Act imposes synthetic-content distribution obligations on very large online platforms (VLOPs); the 2024 revised Product Liability Directive links at the defect-liability layer.

Obligor scope (art. 50)

Section titled “Obligor scope (art. 50)”
  • Provider: the provider of a generative AI system placed on the EU market.
  • Deployer: an entity deploying deepfakes or AI-generated text on matters of public interest.
  • Disclosure to natural persons: when the system interacts with a natural person, that person must be informed that they are “interacting with an AI system” (unless evident from context).

Core obligations

Section titled “Core obligations”
  1. Providers must: mark machine-generated / synthetic audio, image, video, or text output in a machine-readable way, and the marking must be “effective, interoperable, robust, and reliable” (art. 50(2)).
  2. Deployers must:
    • When deploying deepfakes, disclose that the content is AI-generated or manipulated (art. 50(4)); works of art or satire qualify for an exception but disclosure must not disturb appreciation of the work.
    • When AI-generated text is used on matters of public interest, disclose its synthetic nature unless it has undergone human review and a person has taken editorial responsibility.
  3. Emotion recognition and biometric categorisation carry separate disclosure obligations (art. 50(3)).

Final implementation code (June 2026)

Section titled “Final implementation code (June 2026)”

The final voluntary code translates article 50 into practical measures for machine-readable marking, deployer disclosures, chatbot notices and common EU visual icons. Signing does not replace the AI Act and does not immunise a provider or deployer from enforcement; non-signatories may demonstrate compliance through other adequate means.

Exemptions and exceptions

Section titled “Exemptions and exceptions”
  • Partial exemption for law-enforcement use in detection, prevention, investigation, or prosecution of crime.
  • Manifestly artistic, fictional, satirical, or analogous creative works receive weakened disclosure obligations for deepfakes (not eliminated, but in a form that does not disturb artistic appreciation).

Enforcement bodies and penalties

Section titled “Enforcement bodies and penalties”
  • Market surveillance: member-state-designated market surveillance authorities.
  • Coordination: the AI Office (within the Commission) and the European AI Board.
  • Maximum penalties: for breaches of transparency duties, fines of up to €15,000,000 or 3% of global annual turnover, whichever is higher (art. 99).