On 10 June 2026 the European Commission published the final Code of Practice on marking and labelling of AI-generated content, a voluntary tool for meeting the AI-generated content labelling rules in Article 50 of the EU AI Act, which apply from 2 August 2026. Providers of generative AI must mark outputs in a machine-readable way; deployers must label deepfakes and certain AI-written texts. For brand owners and creative businesses, those marks also become evidence when content is copied or manipulated.
Update (October 2026): the Commission now states that it and the AI Board have confirmed the code as an adequate tool to demonstrate compliance, with about 190 signatories by the end of July. The Digital Omnibus on AI, Regulation (EU) 2026/1744, published on 24 July 2026, kept the grace period until 2 December 2026 for systems placed on the market before 2 August 2026. European Commission (31 July 2026); Cuatrecasas (24 July 2026).
Key takeaways
- Article 50 of the AI Act applies from 2 August 2026; systems already on the market have until 2 December 2026, according to the Commission’s FAQ.
- Signatory providers commit to two layers of machine-readable marking: signed metadata and an imperceptible watermark.
- Providers must offer a detection tool, free of charge save for a narrow exception.
- Marking shows that content is synthetic. It does not show who owns the rights.
- Breaching Article 50 can cost up to EUR 15 million or 3% of worldwide annual turnover.
What did the Commission publish?
The Commission announced a voluntary code drafted by independent experts. According to its questions and answers, over 187 participants took part from September 2025. The code has a section for providers (Article 50(2)), a section for deployers (Article 50(4)) and an optional EU icon for labels. It adds no new obligations, but gives signatories a recognised way to show compliance; others must prove their own measures are adequate to each market surveillance authority.
Who must do what under the EU rules on AI-generated content labelling?
| Who | Obligation (AI Act) | How the code applies it |
|---|---|---|
| Providers of generative AI (audio, image, video, text) | Mark outputs in a machine-readable, detectable format (Art. 50(2)) | Signed metadata plus watermark; detection tool; no circumvention tools |
| Deployers publishing deepfakes | Disclose that content is generated or manipulated (Art. 50(4)) | Visible icon or label at first exposure; lighter regime for evidently artistic or satirical works |
| Deployers publishing AI text on matters of public interest | Disclose it, unless there is human review and editorial responsibility (Art. 50(4)) | Labelling rules and editorial control processes |
The provider duty does not apply where the AI only assists standard editing or does not substantially alter the input (Article 50(2)).
How does machine-readable marking work?
The code accepts that no single technique is yet effective, interoperable, robust and reliable on its own, so signatory providers commit to layers:
- Signed metadata: a digitally signed, time-stamped record stating that the content is AI-generated or manipulated.
- Imperceptible watermark: embedded in the content itself, including free-form text longer than 200 tokens.
- Optional fingerprinting or logging, never sufficient alone, and optional provenance data such as system name, provider and timestamp.
- Non-removal: terms of use must ban stripping the marks, and signatories must not promote tools to circumvent them.
Why does marking matter for your IP?
In practice, it gives rightsholders three things. First, evidence: if a manipulated image of your product or spokesperson circulates, a detection tool can help show it is synthetic, supporting takedowns and claims for trademark infringement, unfair competition or image rights. Second, scrutiny: if a competitor claims copyright in an image carrying an AI watermark, it is worth testing how much human authorship it really contains. Third, provenance for your own output.
The limit is clear: a watermark says “made with AI”, not “owned by X”. Where your own metadata identifies the work, author or licence terms, it may also qualify as rights-management information; in Spain, Article 198 of the Intellectual Property Act allows action against anyone who knowingly removes or alters it.
What this means for your business
- Map your role: provider, deployer or both.
- Ask AI vendors which marking techniques they use, whether they will sign and where their detection tool is.
- Check that your publishing workflow does not strip metadata.
- Update agency, influencer and licence contracts to cover AI use, labelling and provenance records.
- Capture detection results whenever you find suspect content.
Our AI and digital assets IP team can review how marking, labelling and ownership fit together, and META Channel’s AI Act practice, within the same group, covers the regulatory side.
Where companies get this wrong
- Treating a watermark as proof of ownership: ownership still depends on authorship, contracts and registrations.
- Losing metadata through compression or content management systems, and with it the evidence.
- Assuming artistic or satirical works need no disclosure: the lighter regime still requires it.
- Waiting too long on misuse. Evidence should be captured before content changes; our IP enforcement team can turn it into action.
Frequently asked questions
Is the EU code on AI-generated content mandatory?
No. Signing the code is voluntary, but the transparency rules in Article 50 of the AI Act are legal obligations from 2 August 2026. Signatories get a recognised way to demonstrate compliance; others must show case by case that their own measures are adequate.
Does an AI watermark prove who owns the content?
No. Marking shows that content was generated or manipulated by AI and, where recorded, which system produced it and when. Ownership depends on human authorship, contracts and, for trademarks or designs, registration. Marking is supporting evidence, not a title.
Can IP Global Guard help us use AI content safely across several markets?
Yes. We review ownership of AI-assisted content, licences and vendor terms, and build evidence routines for enforcement. We act before EUIPO, OEPM and WIPO directly where our professionals are entitled and otherwise through qualified representatives, coordinating local correspondents in Latin America and Africa.
Talk to IP Global Guard about your AI content
Labelling is a compliance task, but the marks it creates are also an IP asset. IP Global Guard, the IP services line of META Channel Corporation Limited, protects AI-assisted content and brands with one strategy and one billing relationship across more than 25 jurisdictions; see our coverage in Europe, Latin America and Africa.
Tell us which AI tools you use, what you publish and where. We will show you where marking, labelling and ownership need attention and coordinate the follow-up from a single point of contact. Contact our AI and IP team.
This article is general information, not legal advice, and reflects the position on its publication date.
Sources
- European Commission, Code of Practice on AI-generated content published (10 June 2026)
- European Commission, Code of Practice text (PDF, 10 June 2026)
- European Commission, Questions and answers on the code (10 June 2026)
- AI Act, Article 50
- AI Act, Article 99 (penalties)
- BOE, Spanish Intellectual Property Act, Article 198
- European Commission, code page (updated 31 July 2026)
- Cuatrecasas, Digital Omnibus on AI published (24 July 2026)
