EU Publishes Final Code of Practice on Transparency of AI-Generated Content

Electric red tourist train in Switzerland,Europe

On June 10, 2026, the European Commission released the final version of the Code of Practice on Transparency of AI-Generated Content. This voluntary framework prepares organizations for the binding transparency obligations under Article 50 of the AI Act, which will become enforceable on August 2, 2026.

Overview

The Code of Practice aims to ensure transparency in AI-generated content, including audio, images, videos, and texts. It is divided into two sections:

• Developers of generative AI systems.
• Deployers (users publishing AI-generated content).

Signatories of the code will demonstrate compliance with legal obligations once the Commission and the AI Committee validate its adequacy.

Key Obligations from August 2, 2026

The AI Act introduces three primary obligations:

1. Machine-Readable Labeling: AI-generated content must include metadata or watermarks that are readable by machines.
2. Clear Labeling of Deepfakes and Public Interest Texts: Deepfakes and AI-generated texts on public interest topics must be clearly labeled unless they undergo human editorial review.
3. Chatbot Transparency: Users must be informed when interacting with a chatbot.

The Code provides technical guidance on implementing these obligations.

Requirements for Providers

Providers of generative AI systems must adhere to four commitments:

1. Watermarking and Metadata:

   • Use at least two overlapping techniques, such as digitally signed metadata (with timestamps) and imperceptible watermarks.
   • Watermarking is mandatory for free-form texts longer than 200 tokens.

2. Detection Mechanisms:

   • Provide free detection tools for users.
   • Providers with fewer than 1 million monthly active users may charge for access, but it must remain free for market surveillance authorities.
   • Detection systems must be interoperable by February 2, 2027, based on open standards or public APIs.

3. Effectiveness and Reliability:

   • Ensure solutions are effective, reliable, and robust.

4. Cooperation with Authorities:

   • Collaborate with national authorities to ensure compliance.

Requirements for Deployers

Deployers must comply with the following:

1. Labeling Deepfakes:

   • Deepfakes (realistic audio, video, or images that appear authentic) must be clearly labeled using standardized EU icons (e.g., "AI generated," "AI modified," or a basic "AI" logo in a circle).

2. Labeling AI-Generated Texts:

   • AI-generated or manipulated texts on public interest topics must be labeled unless reviewed by a human editor.

Timeline

• The Code is now open for signatures.
• Organizations have until August 2, 2026, to align their metadata, watermarking, and labeling processes with the AI Act's requirements.
• The Commission will supplement the Code with additional guidelines to clarify legal obligations and address gaps.

Conclusion

The Code of Practice on Transparency of AI-Generated Content marks a significant step toward ensuring transparency and accountability in AI-generated media. Organizations must act swiftly to comply with the upcoming obligations and avoid penalties under the AI Act.

For further details, refer to the official European Commission documentation.

And the Code of Practice