The EU's chatbot disclosure duty waits until August 2026

The conversation

The European Union's AI Act, formally Regulation (EU) 2024/1689 (cited here through an Internet Archive snapshot, since the live eur-lex page did not load directly during research), entered into force on 1 August 2024 after publication on 12 July 2024. Chapter IV, Article 50, sets a transparency obligation for systems such as chatbots: providers must design a system intended to interact directly with people so those people are informed they are talking to a machine, unless that is already obvious to a reasonably well-informed person given the circumstances.

What the documents show

Article 50 also covers adjacent cases: providers of systems generating synthetic audio, image, video or text must mark outputs as machine-generated where technically feasible, deployers of emotion-recognition or biometric-categorisation systems must tell exposed people, and deployers of systems producing deepfakes or public-interest AI text must disclose that origin, each with its own exception, including a law-enforcement carve-out and, for text, one where a human holds editorial responsibility. Article 50 states this information must reach the person ‘at the latest at the time of the first interaction or exposure.’ Article 113 states that most obligations, including Article 50's, apply from 2 August 2026, two years after entry into force, while a shorter list of chapters is carved out for earlier or later dates.

The system boundary

The regulation frames the boundary as one of awareness rather than capability: a system does not have to refuse a task or hand off to a person to comply with Article 50, it has to make its own machine identity legible at first contact, unless that identity is already obvious. The European Commission's own digital strategy page describes this timeline consistently, stating that transparency obligations for systems such as chatbots take effect alongside the Act's general application date and that the aim is for people to be informed ‘when necessary to preserve trust,’ language that frames disclosure as a floor rather than a substitute for other safeguards.

Where it fails

Because the retrieval date for this entry, 16 September 2026, falls after the 2 August 2026 application date, Article 50's disclosure duty is, on the regulation's own timeline, now in force for systems within scope; a builder should not read the earlier 2024 entry-into-force date, or the manifest's own description of this as a living document, as meaning the duty was optional before this year. What the text leaves to interpretation is the ‘obvious from the circumstances’ exception, which is not itself defined with examples in the operative articles.

• Would a reasonably well-informed, observant person find it obvious, without any disclosure, that they are interacting with an AI system here?
• Is the required disclosure delivered at or before the first interaction, as Article 50(5) specifies, rather than buried later?
• Does the deployment also trigger the separate deepfake or synthetic-content marking duties in Article 50(2) and (4)?

This is a living-document entry: it describes the regulation's text as retrieved on 16 September 2026, not a claim about how any regulator has enforced it in a specific case.