AI Act Code of Practice on Transparency: What You Need to Know

The AI Act Code of Practice on Transparency helps providers and deployers meet Article 50 marking and disclosure duties. Here’s how it works, who it applies…

· 5 min read

What the AI Act Code of Practice on Transparency Is and Why It Matters

If you searched for 'ai act code of practice transparency', you want to know how the EU’s voluntary framework helps organisations comply with Article 50 of the AI Act. The Code of Practice on Transparency of AI-Generated Content, finalised on 10 June 2026 and assessed as adequate by the Commission and the AI Board, offers a practical route for providers and deployers to meet the marking and disclosure obligations under Article 50(2), (4) and (5). It is not law, but signatories can rely on it to demonstrate compliance with those parts of the Act.

Article 50 of Regulation (EU) 2024/1689 sets four transparency duties. The Code focuses on two: the duty for providers of generative AI to mark synthetic audio, image, video and text output in a machine-readable format so it can be detected as artificially generated or manipulated, and the duty for deployers to disclose deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest. It also sets out how the information must be given: clearly, distinguishably, at the latest at first interaction or exposure, and accessibly.

Who the Code of Practice Applies To and What It Covers

The Code is voluntary, but it is designed for two groups: providers of generative AI systems and deployers who publish AI-generated or manipulated content. Providers are expected to implement marking solutions that are effective, interoperable, robust and reliable as far as technically feasible. Deployers must label deepfakes and certain AI-generated text in a way that meets the Article 50(5) requirements.

The Code is split into two sections. Section 1 covers providers, setting out commitments for marking and detection. It calls for a multi-layered approach: metadata that is digitally signed and time-stamped in a secure, tamper-evident way, combined with invisible watermarks embedded in the content itself. Section 2 covers deployers, focusing on the labelling of deepfakes and AI-generated text published to inform the public on matters of public interest.

How the Code of Practice Helps with Article 50 Compliance

For providers, the Code’s Section 1 offers a concrete framework for marking synthetic content. It recommends using open standards like C2PA, which binds cryptographically signed claims about origin and edits to the file. Because metadata can be stripped by screenshots or platforms, the Code also stresses the need for invisible watermarks embedded in the content. This dual approach aligns with the Article 50(2) requirement that marking must be machine-readable and detectable.

For deployers, Section 2 provides guidance on how to disclose deepfakes and AI-generated text. It specifies that labels must be clear, distinguishable and accessible, and that disclosure must happen at the latest at first interaction or exposure. The Code also acknowledges the narrow exemptions in Article 50, such as for business-to-business or industrial contexts, and for artistic or creative works where disclosure must not hamper the display or enjoyment of the work.

Key Steps to Implement the Code of Practice

To use the Code, start by assessing whether you are a provider or deployer under Article 50. Providers should then adopt a multi-layered marking system: use C2PA manifests for images and video, and ensure files carry an invisible watermark. For audio and documents, consider solutions that embed tamper-evident metadata and watermarks. Test your marking to ensure it remains detectable after common transformations like compression or cropping.

Deployers should review their content workflows to identify AI-generated or manipulated material that falls under the disclosure duties. For deepfakes, add a clear label at the point of publication. For AI-generated text published to inform the public on matters of public interest, disclose the use of AI in a way that meets the Article 50(5) requirements. Use the EU’s voluntary icons for labelling where appropriate, and ensure the disclosure is accessible and distinguishable.

What the Code of Practice Does Not Cover

The Code does not address all Article 50 duties. It does not cover the duty for providers to tell people when they are interacting with an AI system, nor the duty for deployers of emotion recognition or biometric categorisation systems to inform exposed individuals. These obligations must be met separately, following the general rules in Article 50(5).

The Code also does not change the scope of Article 50. Content generated and published before 2 August 2026 does not need to be labelled retroactively. The Digital Omnibus on AI, Regulation (EU) 2026/1744, gives generative AI systems placed on the market before 2 August 2026 until 2 December 2026 to meet the marking duty, but this is the only extension. Other deadlines, such as those for high-risk systems, remain unchanged.

How the Code of Practice Fits with Other EU AI Act Guidance

The Code of Practice is one of several tools to help organisations comply with Article 50. The Commission also published non-binding Guidelines on Article 50 on 20 July 2026, which provide further interpretation of the transparency obligations. While the Guidelines are useful for understanding the legal requirements, the Code of Practice offers concrete, actionable steps for providers and deployers.

The EU’s voluntary icons for labelling AI content can be used alongside the Code. The three designs—fully AI-generated, partially AI-modified, and general AI involvement—help deployers meet the disclosure duties in a visually consistent way. However, the icons alone are not sufficient; the underlying marking must still meet the technical requirements of Article 50(2).

What Happens If You Do Not Follow Article 50

Breaching Article 50 can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. Lower ceilings apply to SMEs. While the Code of Practice is voluntary, relying on it can help demonstrate compliance and reduce the risk of enforcement action. However, it does not guarantee legal compliance, and organisations must still ensure they meet all Article 50 obligations.

Next Steps for Providers and Deployers

Providers should review their generative AI systems to ensure they can mark synthetic content in a machine-readable, detectable way. Adopt a multi-layered approach combining metadata and watermarks, and test for robustness. Consider signing the Code of Practice to signal your commitment to transparency.

Deployers should audit their content pipelines to identify AI-generated or manipulated material that requires disclosure. Implement clear, accessible labels for deepfakes and AI-generated text, and ensure disclosures meet the Article 50(5) requirements. Use the EU’s voluntary icons where helpful, but ensure the underlying technical marking is in place.

Frequently asked questions

Is the AI Act Code of Practice on Transparency mandatory?

No, the Code of Practice is voluntary. However, signatories can rely on it to demonstrate compliance with Article 50(2), (4) and (5) of the AI Act.

Who needs to follow the Code of Practice?

The Code is designed for providers of generative AI systems and deployers who publish AI-generated or manipulated content. Providers must mark synthetic output, and deployers must disclose deepfakes and certain AI-generated text.

Does the Code of Practice cover all Article 50 duties?

No, it focuses on the marking duty for providers and the disclosure duties for deployers. It does not cover the duty to inform people when they are interacting with an AI system or the duty for deployers of emotion recognition or biometric categorisation systems to inform exposed individuals.

What marking solutions does the Code of Practice recommend?

The Code recommends a multi-layered approach: metadata that is digitally signed and time-stamped, combined with invisible watermarks embedded in the content. Providers should use open standards like C2PA where possible.

Are there any exemptions from the marking duty?

Yes, there is a narrow exemption for outputs used in business-to-business or industrial contexts. Artistic, creative, satirical or fictional works only need disclosure in a way that does not hamper the display or enjoyment of the work.

Put provenance on your next file

L'AItelier signs images and video with C2PA, adds an invisible watermark, and issues a certificate anyone can check. Your file is deleted as soon as the certificate is issued.

General information about the law as it stood when this was published, not legal advice for your situation.