Policy
EU says voluntary AI transparency code can help with compliance
EU officials said a voluntary code can help companies follow AI transparency rules. Signing it does not prove that a company complies.
European Commission
What happened
The European Commission reviewed the code on July 8, 2026. It found that the code adequately covers several transparency duties in Article 50 of the EU AI Act. The EU AI Board adopted its own assessment the next day.
The code covers labels that software can read, notices for deepfakes, and disclosures for some AI-written public-interest text. The Commission invited AI providers and users to sign it. Signing is voluntary, and regulators can still check whether a company follows the law in practice.
The AI Act requires detectable marking for synthetic audio, images, video, and text when it is technically possible. It also requires disclosure of deepfakes and certain AI-written public-interest material, with specific exceptions.
Why it matters
The code turns vague promises about labeling AI content into specific practices. It asks whether software can detect the label, whether people are told about deepfakes, and whether companies can show how they followed the rules. Those details matter more than simply claiming that content is watermarked.
What remains unclear
- Signing the code is not final proof of compliance. Regulators can still examine what a company actually did.
- Labels and detection tools can fail when watermarks or file information are removed or changed.
- The assessment does not prove that one labeling product works for every kind of AI-generated content. The code may be updated as technology changes.
Sources
-
Commission Opinion on the assessment of the Code of Practice on Transparency of AI-generated content
Used for: The Commission's July 8 conclusion, the AI Board's July 9 adequacy assessment, the voluntary status of the code, its EU-wide scope, and the statement that adherence is not conclusive evidence of compliance.
-
Regulation - EU - 2024/1689 - EN - EUR-Lex
Used for: The text of Article 50's requirements for machine-readable marking of AI-generated content, disclosure of deepfakes, and disclosure of certain public-interest text.
-
European AI Office releases Code of Practice on Transparency of AI-Generated Content
Used for: Independent confirmation that the code addresses provider metadata and marking approaches for AI-generated content.