The least glamorous part of synthetic audio is the part that will matter in the logbook. A station can debate whether listeners notice an AI voiceover, but the EU AI Act asks a duller question: did the publisher disclose it when the content was designed to look or sound real? For radio teams, that means the compliance work happens before broadcast, in scripts, metadata, vendor handoffs and approval notes. The detector debate can continue elsewhere, preferably far from the playout system. ## The deadline is a production date, not a policy mood ArtificialIntelligenceAct.eu says Article 50 transparency obligations apply from 2 August 2026 and are not limited to high risk AI systems. The same source says Article 50 covers four situations, including direct interaction with AI systems and cases where content is AI generated. Wilson Sonsini reports that the European Commission published a Code of Practice on 10 June 2026 for marking and labelling AI generated content, with one section for providers of generative AI systems and another for deployers of AI tools. That matters because a radio publisher is often the deployer, while its voice, editing or ad creation vendor may be the provider. There is a vendor wrinkle worth catching early. ArtificialIntelligenceAct.eu says the AI Omnibus provisional agreement of May 2026 gives generative AI systems already on the market before 2 August 2026 until 2 December 2026 to meet the machine-readable marking requirement under Article 50(2). Translation: your on air disclosure duty may arrive before a supplier has finished its machine marking implementation. Put that in procurement language now, not in a post incident email later. ## What audio publishers actually have to label The Guardian reports that artificially generated images, audio and text designed to look authentic must be labelled, and that companies must ensure people know when they are interacting with artificially generated content designed to look real. Davis+Gilbert puts the obligation in legal terms: beginning August 2, 2026, the EU AI Act will require disclosures for realistic AI generated or AI manipulated depictions of people, objects, places, entities or events that falsely appear authentic or truthful. For radio, that points to synthetic voices, altered speech, reconstructed interview audio and promotional clips that could be mistaken for an authentic recording. Paul Weiss describes the split between provider and deployer obligations under Article 50. Providers must tell users when they are dealing with an AI system and mark AI generated content so it can be detected as such, while deployers must give notice for certain sensitive uses and label deepfakes and AI generated text published on matters of public interest. The practical test is not whether a tool touched the file. It is whether the final content creates an authentic looking or authentic sounding impression that needs disclosure. ## The workflow change is boring, which is the point Wilson Sonsini says the Code is voluntary, but signatories can use it to create a presumption of compliance, while companies taking another route bear the burden of showing equivalent compliance. ComplexDiscovery describes the operational kit around the Code as signed metadata, watermarking, free detection tools and common EU labeling icons. In a radio operation, that means disclosure cannot live only in a presenter note. It needs to survive export, distribution, syndication, web publishing and archive retrieval. Davis+Gilbert also flags a useful boundary for editorial teams that publish web stories, ad copy, show notes or public interest explainers alongside audio. Its alert says AI generated or AI manipulated text used to inform the public on matters of public interest can require disclosure, but meaningful human review and editorial responsibility can eliminate the need for disclosure. That is not a magic phrase for the compliance binder. It means assigning review ownership, recording what changed, and keeping enough provenance to show that a human editor did more than press approve. ## The near term checklist for stations Paul Weiss says the Article 50 transparency obligations broadly take effect from 2 August 2026 and that the Commission published finalised guidelines on 20 July 2026. INSIGHT EU Monitoring similarly reports that the Code is voluntary and sets out practical steps to help providers and deployers meet AI Act transparency obligations that will apply from 2 August 2026. The sensible move for audio teams is to map content types now: live AI chatbot interactions, synthetic voice spots, manipulated clips, AI generated show text and public interest material. Each category needs a disclosure location, a metadata rule, an approval owner and a vendor clause. The penalty backdrop is not theoretical enough to ignore. ComplexDiscovery reports that fines for getting disclosure wrong reach 15 million euros or 3 percent of worldwide turnover. Still, the more immediate risk for publishers is operational: labels omitted during a rush edit, provenance stripped by a distribution tool, or a vendor contract that promises AI features but not marking support. By 2 August 2026, the better run stations will treat AI labels like rights clearances and ad logs, ordinary paperwork that prevents extraordinary meetings. For builders serving radio and audio publishers, the product lesson is plain. Do not sell AI audio features as if disclosure is someone else’s footer problem. The useful tools will attach labels at creation, preserve machine-readable marking through exports, and give editors a review trail they can explain without calling outside counsel every Thursday. ## Sources - The EU AI Act's Transparency Rules: A Practical Guide to Article 50

Sources