On 2 August 2026 the transparency rules in Article 50 of the EU AI Act started to apply: chatbots have to say they are AI, generative systems have to mark their output in a machine readable way, and deepfakes have to be labelled. That date did not move. What did move is the heavier regime for high risk AI, which the Digital Omnibus, Regulation (EU) 2026/1744, pushed back to 2 December 2027 and 2 August 2028.

For anyone whose AI reaches people in Europe, the result is a split calendar. Here is what applies now, what was postponed, what was added, and a short checklist.

What applies since 2 August 2026: four Article 50 duties

As summarised by the law firm Jones Walker and the explainer site artificialintelligenceact.eu:

  • AI interaction, Article 50(1). Providers of chatbots, virtual assistants and other systems that talk directly with people must tell users they are interacting with an AI, unless that is obvious to a reasonably well informed person.
  • Machine readable marking, Article 50(2). Providers of systems that generate synthetic audio, images, video or text must mark the output as artificially generated or manipulated, to the extent technically feasible. Spelling and grammar correction does not count.
  • Emotion recognition and biometric categorisation, Article 50(3). Deployers must inform the people exposed.
  • Deepfakes and public interest text, Article 50(4). Deployers must disclose deepfakes, and AI generated text published to inform the public on matters of public interest, unless a human reviewed it and someone holds editorial responsibility. Evidently artistic, satirical or fictional works get a lighter duty.

The notice must come "at the latest at the time of the first interaction or exposure". The rules also reach companies abroad: the Act applies to providers placing AI systems on the EU market "regardless of where the provider is established", Jones Walker notes, and to providers and deployers whose output is used in the Union.

The broader Article 50 transparency obligations, including the requirement to disclose to users when they are interacting with AI systems, remain unaffected and proceed as scheduled from 2 August 2026.
Gibson Dunn client alert, 27 May 2026

The only transparency extension

Providers of generative systems placed on the market before 2 August 2026 have until 2 December 2026 to add the technical marking. The extension covers only that marking duty, and only those older systems.

Two documents explain how to comply. On 10 June 2026 the European Commission published a final, voluntary Code of Practice on marking and labelling AI generated content. According to Jones Day, it asks for at least two layers of machine readable marking where necessary, such as metadata and watermarks, plus detection tools, and an EU icon or equivalent label for deepfakes, or an audio disclaimer for audio only content. On 20 July the Commission adopted guidelines on Article 50, which are interpretive and do not change the law.

What the Digital Omnibus moved, and what it added

  • High risk systems in Annex III, such as employment, creditworthiness, education and essential services: from 2 August 2026 to 2 December 2027.
  • High risk AI in regulated products, Annex I: from 2 August 2027 to 2 August 2028.
  • National AI regulatory sandboxes: deadline moved to 2 August 2027.
  • A new Article 5 ban on AI that generates child sexual abuse material or non consensual intimate images, video or audio of an identifiable person, covering both sale and use. Companies have until 2 December 2026 to comply.
  • A stronger AI Office with exclusive supervision of certain systems built on general purpose AI models, inspection powers and periodic penalty payments of up to 5 percent of average daily turnover, according to K&L Gates.
  • Smaller changes: systems their providers assess as not high risk must still be registered, with a lighter procedure (Freshfields), and the AI literacy duty is softened from guaranteeing literacy to supporting it (Gibson Dunn).

How the change became law

Negotiators reached a provisional deal in May. The European Parliament approved it on 16 June 2026 with 423 votes in favour, 57 against and 174 abstentions, Eunews reported, and the Council gave its final green light on 29 June. Published in the Official Journal on 24 July, the regulation entered into force on 27 July (nicfab), but each change keeps its own start date.

Supporters call it simplification. "With today's adoption, we are taking another decisive step towards a more competitive European Union," said Marilena Raouna, Cyprus's Deputy Minister for European Affairs, in the Council's announcement. Michael McNamara, one of Parliament's rapporteurs, said "the total ban on these applications will come into force before the end of this year", referring to nudifier apps. The civil liberties group Liberties disagreed, writing that "some of the most worrisome high-risk systems will not be subject to all foreseen protections until 2027 or 2028", and objected that the package passed without an impact assessment.

Fines for transparency breaches

EUR 15 million or 3%
Maximum fine for breaching Article 50, whichever is higher, measured against total worldwide annual turnover. For qualifying SMEs and start ups, the lower of the two applies.

The ceilings come from Jones Walker's reading of the Act. Following the Code of Practice helps, but it is not a presumption of lawfulness: the final assessment rests with the competent authorities, nicfab notes.

A practical checklist for chatbots, voice bots and generated media

  1. Map systems and roles: provider or deployer, and where the output reaches people in the EU.
  2. Chatbots and voice bots: disclose at the start. A visible line in the chat window, a spoken sentence at the start of a call.
  3. Generators: mark the output. Layer metadata and watermarks and offer detection. Systems on sale before 2 August have until 2 December 2026; newer ones must comply now.
  4. Published media: label deepfakes visibly, and with an audio disclaimer for audio only content.
  5. AI written news: keep editorial control, and document the human review.
  6. Write down your exceptions, such as obvious AI use or satire.
  7. Add intimate imagery safeguards to image, video and audio generators before 2 December 2026.
  8. Keep the high risk work moving. Annex III duties are delayed to 2 December 2027, not cancelled.