Article 50, answered straight.

Yes, unless it is already obvious. Art. 50(1) requires that people interacting with an AI system are informed they are interacting with AI, and Art. 50(5) says that information must reach them at the latest at the time of the first interaction, in a clear and distinguishable manner. The duty formally sits on the provider of the chatbot, but you are the business the customer sees, so in practice you want the notice visible on every channel the assistant runs on.

The "obvious" exemption is judged from the point of view of a reasonably well-informed person — your average customer, not you. The full guide to chatbot disclosure walks through what counts.

Possibly not — Art. 50(1) does not apply where it is obvious to a reasonably well-informed, observant and circumspect person that they are dealing with AI. But the bar is higher than most people assume, and the person judging it afterwards is not you.

Our position, and we say it against our own interest: a one-line notice costs nothing and removes the argument entirely. Why the exemption is thinner than it looks.

No. The operative words are in Art. 50(5): the information must be provided "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure". A clause inside a document nobody opens is neither clear at first interaction nor distinguishable — it is the opposite of both.

The underlying duty for a chatbot is Art. 50(1), and it is about the moment the conversation starts. Put the disclosure where the interaction happens: in the chat window, in the greeting, on the content. More on placement per channel.

No. The AI Act prescribes no template and no official sentence. What it sets is a standard, in Art. 50(5): clear, distinguishable, and no later than the first interaction or exposure. Anything that meets it is compliant wording; anything that does not is not, however official it sounds.

The Commission's Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 (published 20 July 2026) give practical examples and definitions. They are interpretive guidance from the Commission — the binding text is the Regulation itself. Article 50, duty by duty.

It depends on what the image shows, and there are two separate layers. The visible label is your duty as a deployer under Art. 50(4), and it bites when the content appreciably resembles real people, places or events and could be taken for authentic. A purely fantastical illustration is a different case.

The machine-readable mark inside the file is a different duty on a different party — see the next answer. The full guide, and our free EU AI-label icon picker if you just need the label.

Your AI tool's provider. Art. 50(2) puts the duty to mark synthetic audio, image, video and text in a machine-readable format on providers of the AI system that generates it, not on the business publishing the output.

Two things are still yours: choosing tools that actually mark, and not stripping the marks afterwards — plenty of content management systems quietly discard image metadata on upload. Marking has applied since 2 August 2026, and AI tools already on the market before that date have a transitional period until 2 December 2026 (Reg. (EU) 2026/1744). What to check, tool by tool.

The Act's definition has nothing to do with intent. It covers AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic (Art. 3(60)). An AI video of your own shop qualifies. So does a cloned voice-over of your own voice.

Where the work is evidently artistic, creative or satirical, Art. 50(4) limits the disclosure to something that does not spoil the enjoyment of the work — a credit line rather than a banner across it. The full guide.

Yes. The duty follows the AI system, not the website. An assistant answering on WhatsApp, Messenger or a phone line is the same Art. 50(1) interaction, and Art. 50(5) still wants the disclosure at the first interaction — which on a phone line means spoken, first, not buried in a menu.

How to disclose per channel.

Partly, and the part that applies is smaller than the headlines suggest. Most small businesses are deployers — they use AI tools rather than build them — and the transparency duties of Art. 50 are the ones that realistically land on them. The heavy high-risk obligations generally do not.

Size does not create an exemption from Art. 50. It does affect penalties — see below. What actually hits an SME.

For breaches of the transparency obligations, Art. 99(4)(g) sets fines of up to €15,000,000 or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher.

Read it together with Art. 99(6): where the offender is an SME, including a start-up, each of those ceilings applies as whichever is lower. That is the provision that turns an unpayable headline number into a proportionate one for a small business — and it is the part most scare-marketing leaves out. Fines are set by national market surveillance authorities, which also weigh the size of the operator and the nature of the breach.

The numbers without the fear-mongering.

No, and this is the single most common misreading we see. Reg. (EU) 2026/1744 — the "Digital Omnibus on AI" — pushed back deadlines in the high-risk part of the AI Act. It did not exclude Article 50, and it did not postpone Article 50.

The one piece of relief it gives to Article 50 is narrow: a transitional period for the Art. 50(2) machine-readable marking duty, running to 2 December 2026, and only for AI systems that were placed on the market before 2 August 2026. Everything else in Article 50 has applied since 2 August 2026.

Official text: Reg. (EU) 2026/1744 on EUR-Lex. Our dated log of what changed: the changelog.

Two. 2 August 2026: Article 50 applies — chatbot disclosure, deepfake and public-interest-text disclosure, and machine-readable marking. 2 December 2026: the end of the transitional period for Art. 50(2) marking, for AI tools that were already on the market before 2 August 2026.

If you are a business using AI tools rather than building them, the second date is your suppliers' deadline, not yours. We track both, dated and sourced, in the changelog and on the Article 50 radar.

The Code of Practice on Transparency of AI-generated Content is a voluntary instrument, published 10 June 2026, aimed mainly at the providers of AI systems that generate content. The Commission and the AI Board have assessed it as an adequate voluntary tool for demonstrating compliance with the AI Act's transparency obligations.

Voluntary is the operative word: not signing it breaks no law, and signing it does not replace Article 50. As the Commission puts it, adherence to the code is voluntary while the Article 50 requirements are legal obligations. The Code on the Commission's site.

Not for Article 50, and not from us. No private company can certify you as compliant with a regulation — only a competent authority's own assessment counts, and for these transparency duties there is no certification scheme at all.

What can honestly be sold is implementation: the disclosure actually installed, the labels actually written, and a dated record of what was done. That is what our Kit is, and why it is described as a technical implementation tool rather than a compliance product.