EU AI transparency Code vs Article 50 guidelines: what each does

The Code of Practice is voluntary and covers Article 50(2), (4) and (5); the guidelines clarify scope for all of Article 50. How the pages split them.

4 min readSume
All posts

They are two documents with different jobs. The Code of Practice on Transparency of AI-generated Content is voluntary and sets measures for Article 50(2), (4) and (5), the marking and labelling duties. The Guidelines on transparency obligations clarify the scope of the transparency obligations, and the Commission describes them as complementing the code. Neither replaces the AI Act text. This reports the Commission's pages and is not legal advice.

What does the Code of Practice cover?

The code page says it supports compliance with Article 50(2), (4) and (5) and has two sections: Section 1 for providers, with rules for marking and detection of AI-generated and manipulated content, and Section 2 for deployers, with rules for labelling deepfakes and AI-generated and manipulated text. Adherence is voluntary, but the page is explicit that the Article 50 requirements are legal obligations.

The Commission and the AI Board have confirmed the code as an adequate voluntary tool to show compliance.

What do the guidelines cover?

The guidelines page says they help providers, deployers and competent authorities comply with Article 50, clarify the obligations and give legal certainty about scope. The FAQ built on them explains concepts such as the four criteria for a two-way AI interaction, the three criteria for a deepfake, and what counts as human review of text.

For Article 50(1) and (3), the FAQ says providers and deployers can determine adequate compliance measures themselves while taking the guidelines into account. There is no code section for those two paragraphs.

How the Commission pages split the work, read 2026-10-02
Article 50 paragraphDuty in shortWhere the Commission points
50(1)Tell people they are interacting with AIGuidelines; choose your own measures
50(2)Machine-readable marking and detection (providers)Code Section 1, plus guidelines on scope
50(3)Inform people exposed to emotion recognition or biometric categorisationGuidelines; choose your own measures
50(4)Label deepfakes and certain AI text (deployers)Code Section 2, plus guidelines on scope
50(5)Information provided to the people concernedCode, cross-cutting requirements

Which one do I read first?

Start with the guidelines or the FAQ to decide whether your output is in scope at all, for example whether a clip is a deepfake or an exempt standard edit. Then read the code section that matches your role if you plan to rely on it. The code page says signatories can rely on its measures to demonstrate compliance, while others must show their own measures are adequate.

For video teams, the follow-on question is which of your files are covered. The Article 50 evidence log post covers keeping a record per file.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume