Code of Practice on AI-Generated Content: What It Means for You

The EU Code of Practice on Transparency of AI-Generated Content sets clear rules for providers and deployers. Learn its two sections, commitments, and why…

· 4 min read

What the Code of Practice on AI-Generated Content Is and Why It Exists

The Code of Practice on Transparency of AI-Generated Content is a voluntary framework finalised on 10 June 2026 to help organisations meet the EU AI Act’s Article 50 marking and disclosure obligations. It was assessed as adequate by the European Commission and the AI Board, meaning signatories can rely on it to demonstrate compliance with Article 50(2), (4) and (5). The Code does not replace the law—it offers a practical, industry-backed way to implement the transparency rules that apply from 2 August 2026.

Article 50 of the EU AI Act requires providers of generative AI systems to mark synthetic audio, image, video and text in a machine-readable format so it can be detected as artificially generated or manipulated. Deployers must disclose deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest. The Code translates these duties into actionable steps, with Section 1 for providers and Section 2 for deployers.

Section 1: What Providers of Generative AI Systems Commit To

Section 1 applies to providers of generative AI systems—those who develop or place such systems on the market. It focuses on the marking duty under Article 50(2), which must be met by embedding machine-readable information in the output. The Code calls for a multi-layered approach: visible labels, machine-readable metadata, and invisible watermarks where technically feasible.

Signatories commit to ensuring their systems generate content with metadata that is digitally signed and time-stamped in a secure, tamper-evident way. This means the provenance of the content can be verified even if the file is shared or republished. The metadata must be interoperable, robust and reliable, and should include details such as the model used, the date of generation, and any edits or manipulations applied.

Section 2: What Deployers of AI Systems Commit To

Section 2 applies to deployers—organisations or individuals who use generative AI systems to create or publish content. It covers the disclosure duties under Article 50(4), which require deployers to label deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest.

Deployers who sign the Code commit to clearly and prominently disclosing when content is AI-generated or manipulated, using labels or other methods that meet the Article 50(5) standards: clear, distinguishable, provided at the latest at first interaction or exposure, and accessible. The Code also encourages deployers to use the EU’s voluntary icons for labelling AI content, which distinguish between fully AI-generated, partially AI-modified, and general AI involvement.

Why Signed, Time-Stamped Metadata Matters

Signed, time-stamped metadata is the backbone of the Code’s approach to transparency. Without it, there is no reliable way to prove when, how, or by whom content was generated. Digital signatures ensure the metadata cannot be altered without detection, while time-stamps provide a verifiable record of creation.

This matters because the EU AI Act requires marking solutions to be effective, interoperable, robust and reliable as far as technically feasible. Metadata that is not signed or time-stamped can be stripped or tampered with, making it difficult to verify provenance. For example, screenshots or platform processing can remove metadata, but an invisible watermark embedded in the content itself remains detectable. Combining metadata with watermarks—such as in C2PA manifests or similar systems—addresses this risk.

How the Code of Practice Relates to the EU AI Act

The Code of Practice is not a legal requirement, but it is a practical tool for compliance. Signing it is voluntary, but the Commission’s adequacy assessment means that following its provisions can be used as evidence of compliance with Article 50(2), (4) and (5). This is particularly useful for organisations that want to demonstrate transparency without reinventing the wheel.

The Code also aligns with the Commission’s Guidelines on Article 50, published on 20 July 2026. These non-binding guidelines clarify how to meet the marking and disclosure obligations, including the use of machine-readable formats and clear labelling. The Code complements these guidelines by offering a structured, industry-led approach to implementation.

Practical Steps to Implement the Code’s Requirements

For providers, the first step is to embed machine-readable metadata in all synthetic outputs, including a digital signature and time-stamp. This metadata should include the model used, the date of generation, and any edits. Providers should also consider adding invisible watermarks to ensure detectability even if metadata is removed. Tools like C2PA manifests can help achieve this, as they bind claims about origin and edits to the file in a cryptographically signed format.

For deployers, the focus is on disclosure. When publishing AI-generated or manipulated content, use clear labels that meet the Article 50(5) standards. The EU’s voluntary icons can be a simple way to signal AI involvement. Deployers should also document their processes for identifying and disclosing AI-generated content, especially for matters of public interest.

Exemptions and Special Cases Under Article 50

Not all AI-generated content requires marking or disclosure. There is a narrow exemption for outputs used in business-to-business or industrial contexts, where the content is not intended for public consumption. Artistic, creative, satirical or fictional works only need disclosure in a way that does not hamper the display or enjoyment of the work. Additionally, AI-generated text under human review with editorial responsibility is exempt from the public-interest text disclosure duty.

Content generated and published before 2 August 2026 does not need to be labelled retroactively. However, any new content published after this date must comply with the marking and disclosure obligations, unless it falls under one of the exemptions.

Frequently asked questions

Is signing the Code of Practice on Transparency of AI-Generated Content mandatory?

No, signing the Code is voluntary. However, it has been assessed as adequate by the Commission and the AI Board, so signatories can rely on it to demonstrate compliance with Article 50(2), (4) and (5) of the EU AI Act.

What are the two sections of the Code of Practice?

Section 1 covers providers of generative AI systems and focuses on marking synthetic outputs with machine-readable metadata. Section 2 covers deployers and requires them to disclose deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest.

Why is signed, time-stamped metadata important?

Signed, time-stamped metadata ensures the provenance of AI-generated content can be verified. Digital signatures prevent tampering, while time-stamps provide a verifiable record of creation, making it easier to detect and trace AI-generated or manipulated content.

Do the exemptions in Article 50 apply to all duties?

No, each exemption applies only to the specific duty it names. For example, the exemption for business-to-business or industrial contexts applies only to the marking duty, not to the disclosure duties for deployers.

What happens if I do not comply with Article 50?

Breaching the Article 50 obligations can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. Lower ceilings apply to SMEs.

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.