Guide · Art. 50(7)

EU AI Act Code of Practice: Transparency of AI-Generated Content Explained

Last updated: 8 November 2026

The EU AI Act Code of Practice on transparency of AI-generated content is a voluntary code, published on 10 June 2026, on marking and labelling AI-generated content under Article 50(2) and (4). Adhering to a code assessed as adequate is one way to show compliance, but it gives no presumption of conformity. Article50.io scans pages for visible AI disclosures.

This page is general information about the code and Article 50, not legal advice.

Scan for free

What the Code of Practice on transparency is

The Code of Practice on Transparency of AI-Generated Content is a voluntary code drawn up at EU level. It is about how AI-generated or manipulated content is marked, detected and labelled under Article 50 of the EU AI Act (Regulation (EU) 2024/1689).

Two duties sit behind it. Article 50(2) says providers of AI systems that generate synthetic audio, image, video or text must make sure the outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated”. Article 50(4) says deployers of an AI system that produces deep fakes, or AI text “published with the purpose of informing the public on matters of public interest”, must disclose that the content is artificially generated or manipulated.

The code does not add legal duties; it describes one way of meeting duties that already exist. For the wider picture, see Article 50 explained and our guide to Article 50 transparency obligations.

The legal basis: Article 50(7), before and after the Digital Omnibus

Codes of practice for AI-generated content come from Article 50(7). The Digital Omnibus on AI, Regulation (EU) 2026/1744 of 8 July 2026, replaced that paragraph.

Article 50(7) in the original text and as replaced by the Digital Omnibus
	Original Article 50(7) (2024)	As replaced by the Omnibus (2026)
Who encourages codes	The AI Office	The Commission “shall encourage and facilitate the drawing up of codes of practice at Union level”
Subject	“the detection and labelling of artificially generated or manipulated content”	“the detection, marking and labelling of artificially generated or manipulated content”
What the Commission does with a code	“may adopt implementing acts to approve those codes of practice”	Assesses, “taking utmost account of the opinion of the Board”, whether adherence is “adequate to ensure compliance with the obligations laid down in paragraphs 2 and 4”
If the code falls short	If it deems the code not adequate, the Commission may adopt an implementing act with common rules (Article 98(2) procedure)	Same: if it deems the code inadequate, the Commission may adopt an implementing act with common rules (Article 98(2) procedure)

The Commission's adequacy assessment follows the procedure in Article 56(6); an implementing act would follow the examination procedure in Article 98(2). Recital 41 of the Omnibus is clear about the limits: codes under Article 50(7) “have limited legal effect, and in particular do not grant a presumption of conformity”. Signing the code therefore does not mean an authority must treat you as meeting the law.

Status: published and voluntary

The Commission's guidelines on Article 50 (C(2026) 7702 final, 5 November 2026, non-binding) state that the Code of Practice on Transparency of AI-Generated Content was “published on 10 June 2026” (footnote 43). The 20 July 2026 text (C(2026) 5054 final) was a draft submitted for approval. Signing the code is voluntary; no provider or deployer has to sign.

Article 50 itself is not voluntary. It has applied since 2 August 2026, whether or not you sign anything. The guidelines call themselves a “first interpretation” and say they will be reviewed (para. 155), so the practical picture may change.

What signing means in practice

Paragraphs 146 to 150 of the guidelines set out how the code is used in supervision:

✔Providers and deployers covered by Article 50(2) and (4) “may demonstrate compliance” by adhering to a code “that is assessed as adequate pursuant to Article 50(7)” (para. 146).
✔The Commission calls this “a straightforward, predictable, and legally certain way of demonstrating compliance”. For signatories, authorities “will focus their supervisory activities on assessing whether those signatories have adhered to the code” (para. 147).
✔Opting out of sections of the code loses that benefit for those sections (para. 147).
✔Non-signatories must show compliance “through other adequate means”, for example a gap analysis against the code, and “will likely” face more detailed requests for information (para. 148).
✔Commitments under an adequate code may be taken into account as a mitigating factor when a fine is set, under Article 99(7)(e) (para. 149).
✔If a code is not deemed adequate, the Commission may adopt an implementing act with common rules (para. 150).

Fines are the background here. Under Article 99(4), breaches of Article 50 can lead to fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher; for SMEs, including start-ups, and small mid-caps, whichever is lower (Article 99(6) and (6a)). See Article 50 fines and penalties.

What the code contains

The code has two sections: one for providers, on marking and detection, and one for deployers, on labelling deep fakes and public-interest text. For providers, its Measure 1.1 asks for two layers of machine-readable marking: digitally signed metadata (Sub-measure 1.1.1) and an imperceptible watermark (Sub-measure 1.1.2), with an exception for short text.

In its opinion C(2026) 4839 of 8 July 2026, the Commission concluded that the code adequately covers Article 50(2), (4) and (5); the AI Board adopted its own assessment on 9 July 2026. Germany's Federal Network Agency (Bundesnetzagentur) also explains the code on its Code of Practice page.

Signing the code is not a safe harbour. It is a reference for demonstrating compliance, not a substitute for the Article 50 duty itself: the duty to mark, or to disclose, still applies to the content.

The AI Act itself names no required technique. Recital 133 lists examples such as “watermarks, metadata identifications, cryptographic methods for proving provenance and authenticity of content, logging methods, fingerprints”. More on the techniques in EU AI Act watermarking.

What it means for you

The code is about marking and labelling AI-generated content under Article 50(2) and (4). How much it matters depends on your role.

What the code means for providers, deployers and website owners
Who you are	Duty in the AI Act	Where the code fits
Provider of a generative AI system (image, audio, video or text generator)	Article 50(2): machine-readable marking and detectability of outputs	Signing is one route to showing compliance; otherwise expect to show other adequate means
Deployer publishing deep fakes or public-interest AI text	Article 50(4): disclose that the content is artificially generated or manipulated	Signing is one route to showing compliance; see deepfake disclosure
Website owner using a chatbot	Article 50(1): a provider duty to inform people they are interacting with an AI system	The code is about marking and labelling of AI-generated content (Art. 50(2) and (4))

Generative AI providers. Under Article 3(3), you are a provider if you develop an AI system, or have one developed, and place it on the market or put it into service under your own name or trademark. The guidelines add that a company that modifies an existing generative system and puts it into service under its own name becomes a provider of the new system (para. 11). Marking has to be both machine-readable and detectable; doing one without the other “will not suffice” (paras. 69–70).

Deployers of deep fakes and public-interest text. The text duty does not apply where the text “has undergone a process of human review or editorial control” and someone holds editorial responsibility. For evidently artistic, creative, satirical or fictional work, disclosure is limited to the existence of the content, “in an appropriate manner that does not hamper the display or enjoyment of the work”.

Website owners with a chatbot. Whether a chatbot needs an AI notice is a separate question under Article 50(1); our Article 50 questions page covers it.

Key dates

✔10 June 2026: the code was published (Commission guidelines, footnote 43).
✔8 July 2026: date of the Digital Omnibus on AI, Regulation (EU) 2026/1744, which replaced Article 50(7); published in the Official Journal on 24 July 2026. Same day: Commission opinion C(2026) 4839 finding the code adequate for Article 50(2), (4) and (5).
✔9 July 2026: the AI Board adopted its own assessment of the code.
✔20 July 2026: draft Article 50 guidelines circulated for approval (C(2026) 5054 final). 5 November 2026: the Commission adopted the guidelines as C(2026) 7702 final. They are non-binding (paragraph 5). Adoption is not a change to the Act.
✔2 August 2026: Article 50 applies, all paragraphs.
✔2 December 2026: providers of generative AI systems placed on the market before 2 August 2026 must meet the Article 50(2) marking duty (Article 111(4), inserted by the Omnibus).

Where Article50.io fits

Article50.io does not sign codes for you or check whether an AI provider marks its outputs: that duty belongs to the generator provider and is not observable from a public page. A scan loads one public page in a real browser and reports what it detects: chat widgets, voice agents and AI avatars without an AI disclosure, and up to 8 main images whose metadata (an IPTC digital source type or a C2PA manifest) says they were AI-generated, with nothing on the page saying so. A missing marker is never a finding.

It is not legal advice, and a clean result is not a compliance clearance. Run a free scan to see the most severe finding, or see pricing for the €499 scan report with a ready-to-paste fix for each finding.

Frequently Asked Questions

What is the EU AI Act Code of Practice on transparency?

It is a voluntary code on marking, detecting and labelling AI-generated or manipulated content, linked to Article 50(2) and (4) of the AI Act. The Commission's Article 50 guidelines state it was published on 10 June 2026. Adhering to a code assessed as adequate is one way to demonstrate compliance with those paragraphs.

Is signing the Code of Practice mandatory?

No. Signing is voluntary. The underlying duties in Article 50 are not, and have applied since 2 August 2026. According to the Commission's guidelines, providers and deployers who do not sign must show compliance through other adequate means, such as a gap analysis against the code, and will likely face more detailed requests for information.

Does signing the code give a presumption of conformity?

No. Recital 41 of the Digital Omnibus on AI says codes under Article 50(7) have limited legal effect and in particular do not grant a presumption of conformity. In practice, authorities supervising signatories will focus on whether they adhered to the code, and commitments under an adequate code may count as a mitigating factor when fines are set.

What did the Digital Omnibus change in Article 50(7)?

Originally the AI Office encouraged codes and the Commission could adopt implementing acts to approve them. Now the Commission encourages codes and assesses, taking utmost account of the AI Board's opinion, whether adherence is adequate for paragraphs 2 and 4. In both versions the Commission may adopt an implementing act with common rules if a code is inadequate.

Does the code apply to my website chatbot?

The code is about marking and labelling AI-generated content under Article 50(2) and (4). Telling people they are interacting with an AI system is a separate duty in Article 50(1), which falls on the provider of the AI system. Whether your chatbot needs a notice depends on the facts.

Does my chatbot need an AI disclosure?

What happens if a signatory opts out of parts of the code?

According to paragraph 147 of the Commission's guidelines, opting out of sections loses the benefit of easier demonstration of compliance for those sections. For the parts not followed, the provider or deployer would need to show compliance through other adequate means.

Does content made before 2 August 2026 need marking?

Paragraph 154 of the Commission's guidelines (C(2026) 7702 final, 5 November 2026, non-binding) says outputs within Article 50(2), and deep fakes within Article 50(4), generated or manipulated before 2 August 2026 need not be marked or labelled retroactively. The same applies to public-interest text that was both generated and published before that date. Text generated before 2 August 2026 but first published on or after it needs labelling where Article 50(4) applies. Providers of generative systems already on the market have until 2 December 2026 for Article 50(2) marking only (paragraph 153 and Article 111(4)). That date is not a chatbot or deployer disclosure deadline.

Check your site automatically

Article50.io is an automated scanner for the Article 50 transparency rules of the EU AI Act. It scans a website for potential Article 50 transparency gaps and generates a report of what the scan found and what it could not check, with remediation guidance, implementation instructions and disclosure wording you can paste in.

The free scan shows your single most severe finding. Scans are capped at 45 seconds — no signup, public pages only. The full scan report is €499 one-off, with a ready-to-paste fix for every finding: see pricing.