Draft Council text would force machine‑readable provenance under AI Act Article 50

The European Commission on May 8 published draft guidelines for Article 50 of the EU AI Act that spell out how transparency rules should work when people encounter AI systems or synthetic content. The draft says providers of systems that generate synthetic audio, image, video or text must ensure those outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated. Article 50 itself applies from August 2, 2026 under the AI Act’s timetable. The Commission said the draft guidance is open for consultation until June 3, 2026. ### What exactly does Article 50 require providers to do? Article 50(2) says providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content “shall ensure” that outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The same provision says technical solutions should be effective, interoperable, robust and reliable “as far as this is technically feasible,” taking account of content type, implementation costs and the state of the art. (digital-strategy.ec.europa.eu) The Commission’s code-of-practice page breaks that into operational terms for providers: outputs across audio, image, video and text should be marked in machine-readable form and detectable, with technical measures designed around interoperability and reliability. The code is voluntary if approved, but the underlying Article 50 obligation is in the law. ### Does the rule cover only images and video? (artificialintelligenceact.eu) Article 50(2) in the AI Act covers synthetic audio, image, video and text. The draft-guidelines page published by the Commission describes the document as guidance for competent authorities, providers and deployers on complying with Article 50 consistently across the bloc. The code-of-practice page separates provider duties from deployer duties. Providers are the companies shipping systems that generate content and must handle machine-readable marking and detectability. (digital-strategy.ec.europa.eu) Deployers face separate disclosure duties for deepfakes and for certain AI-generated or AI-manipulated text published on matters of public interest, subject to the exceptions set out in the Act. (artificialintelligenceact.eu) ### Why does this land on product and engineering teams now? The Commission’s text ties compliance to the output itself, not just to a user-facing label. Because Article 50(2) requires outputs to be machine-readable and detectable, providers that sell into Europe will need those markers to survive generation, export and distribution workflows if they want the delivered file or artifact to remain compliant. That is an inference from the legal requirement’s focus on outputs and technical solutions, rather than a phrase the Commission uses in those words. (digital-strategy.ec.europa.eu) Working group 1 in the Commission’s code process is focused on providers and on technical solutions for marking and detection. The Commission also says the process is meant to promote cooperation across the value chain, a signal that compliance work will not sit only with policy teams. ### Is this already final guidance? The May 8 document is draft guidance, not final guidance. The Commission said it prepared the guidelines in parallel with the code of practice on marking and labelling of AI-generated content, and invited stakeholders to take part in a targeted consultation through June 3. (artificialintelligenceact.eu) The legal baseline, however, is already fixed in Regulation (EU) 2024/1689. (digital-strategy.ec.europa.eu) The Official Journal version of the AI Act says the regulation applies from August 2, 2026, with staged exceptions for some other provisions. Article 50 is part of that timetable. ### What happens next before August 2026? June 3, 2026 is the deadline the Commission gave for feedback on the draft Article 50 guidelines. The Commission is also running the code-of-practice process on marking and labelling of AI-generated content through expert working groups covering providers and deployers. (digital-strategy.ec.europa.eu) August 2, 2026 is the date Article 50 applies under the AI Act. (eur-lex.europa.eu) By then, providers placing relevant systems on the EU market will need output-marking and detectability measures in place, while deployers facing deepfake and public-interest disclosure duties will need their own compliance processes ready. (artificialintelligenceact.eu) (digital-strategy.ec.europa.eu)