EU Code of Practice on Transparency of AI-Generated Content
Issued by
European Commission
- September 30, 2026 · Correction — Added the official link to the Commission page. (Cody Maxwell)
- September 30, 2026 · Correction — Corrected who each section covers, the two-technique marking rule, the dates, and removed an invented annual reporting duty and safe-harbor claim (fact-check finding). (Cody Maxwell)
- October 1, 2026 · Correction — Rewrote the practical steps, newsletter hook, search description, and audience fields to match the entry's corrected content. (Cody Maxwell)
The European Commission published this voluntary Code of Practice to support Article 50 compliance under the EU AI Act. It addresses generative AI providers and deployers serving the EU. The Code covers content labeling, controls that record where content came from, and disclosure workflows.
Applies To
Overview
The Code of Practice on Transparency of AI-Generated Content is a voluntary instrument developed by the European Commission to put into practice the transparency requirements set out in Article 50 of the EU AI Act. It provides concrete technical and organizational guidance for generative AI providers and deployers responsible for ensuring that AI-generated content is identifiable by end users. Key provisions address machine-readable provenance signals (data in a file showing where content came from), human-readable disclosure labels, and internal governance processes for maintaining content authenticity records. Signing gives organizations a streamlined way to show compliance with Article 50, though it is not a formal presumption of conformity. Enforcement of the underlying Article 50 obligations remains with national market surveillance authorities and, for general-purpose AI model providers, the European AI Office. The final text was published on 10 June 2026, and the Article 50 duties apply from 2 August 2026. Marking duties for generative systems already on the market apply from 2 December 2026.
Key Requirements
- •Providers of generative AI systems mark outputs with at least two machine-readable techniques, such as signed metadata plus an imperceptible watermark (Section 1).
- •Providers must make their watermark detection interoperable by 2 February 2027.
- •Deployers who generate deep fakes, or publish AI-generated text on matters of public interest, label it visibly (Section 2).
- •Deployers can use an optional EU icon to label AI-generated content.
- •Signatories keep internal compliance processes and documentation; the Code sets no annual report to the AI Office.
- •Article 50 applies from 2 August 2026, and breaches can bring fines of up to EUR 15 million or 3% of global annual turnover.
What Your Organization Must Do
- →Map which of your generative AI systems and uses fall under Article 50 transparency duties.
- →Decide whether to sign the voluntary Code, and record the business reasons either way.
- →Plan two machine-readable marking techniques for outputs, such as signed metadata plus an imperceptible watermark.
- →Set visible labels for deep fakes and AI-generated text published on matters of public interest.
- →Diary 2 August 2026, 2 December 2026 and 2 February 2027 as internal readiness dates.
- →Keep internal documentation of your marking and disclosure processes for market surveillance authorities.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- Does signing the CoP-AIGC create a safe harbor against EU AI Act Article 50 enforcement?
- Not formally; signing gives a streamlined way to show compliance with Article 50, and regulators will focus on whether signatories follow the Code. It is not a legal presumption of conformity.
- What are the maximum fines for violating Article 50 of the EU AI Act that the CoP-AIGC supports?
- Non-compliance with Article 50 can result in fines of up to EUR 15 million or 3% of global annual turnover, whichever is higher. The Code itself is voluntary, but the underlying Article 50 obligations carry these binding penalties.
- Which technical standards satisfy the machine-readable provenance signal requirement under the CoP-AIGC?
- The Code requires at least two machine-readable techniques, such as digitally signed metadata plus imperceptible watermarking. It does not name C2PA or any single standard, and detection must be interoperable by 2 February 2027.
- Does the CoP-AIGC apply to SMBs deploying third-party generative AI tools, or only to model developers?
- Only in part: Section 1 covers providers of generative AI systems. Deployers fall under Section 2 when they generate deep fakes or publish AI-generated text on matters of public interest.
- How often must CoP-AIGC signatories submit self-assessment reports, and to which authority?
- The Code sets no annual self-assessment report to the AI Office. Signatories commit to internal compliance processes and documentation, and market surveillance authorities check compliance.
- When does the CoP-AIGC take effect relative to the broader EU AI Act enforcement timeline for generative AI?
- The Code was published on 10 June 2026, and Article 50 applies from 2 August 2026. Marking for generative systems already on the market applies from 2 December 2026, and detection must be interoperable by 2 February 2027.