What counts as an AI disclosure, according to the Commission
On 20 July 2026 the Commission adopted its final guidelines on Article 50. Paragraphs (37) and (38) are the most practically useful thing published on this subject: one list of techniques that work, one list of techniques that do not. Here they are.
The obligation
Article 50(1) requires providers to ensure that people
are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect…
And Article 50(5) says how:
The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. The information shall conform to the applicable accessibility requirements.
Paragraph 50(5) is the one most summaries skip, and it is the one that actually tells you where to put the thing.
What works
The Commission's guidelines, paragraph (37), on textual disclosure:
Prominent, plain-language labels or banners (e.g. "You are interacting with an AI system") and first-turn greetings in chatbots that may be combined, as appropriate, with persistent badges or labels visible throughout the interaction as proportionate to the specific context and risks. Furthermore, it is recommended to position accompanying disclosures close to the interaction interface (e.g. near the input/output field) and use simplified wording, particularly for users with lower digital literacy and children.
That is a specification. Prominent, plain language, in the chat, near the input field, at the first turn. The example wording is the Commission's own — which is why our widget uses it verbatim.
On substance, paragraph (35): people must be
explicitly informed about the artificial, non-human nature of the interacting counterpart.
What does not work
Paragraph (38), quoted in full because every item is a pattern in the wild:
Certain techniques are not necessary and, when used alone, insufficient for effectively fulfilling the transparency obligation in Article 50(1) and (5) AI Act, notably:
- Disclosures contained only in terms and conditions, URLs, or documentation (such disclosures may complement, though not replace, in-context disclosure);
- Machine-readable markings (e.g. metadata or watermarks) that are not perceivable by users at the point of interaction. […]
- Unclear or ambiguous signals (e.g. generic references to "assistant") or human-like representations that may mislead users;
- Generalised disclosures that are not sufficiently specific to the AI system's outputs and interactions (e.g. on a platform providing a variety of services or individual inputs, a general disclosure like "Services on this website use AI" is insufficient);
- Technical or capability-based descriptions: statements solely referring to underlying technologies (e.g. "this system uses LLMs") without explaining the function or implications of the system for the user and its artificial origin.
And paragraph (142), on what fails the "clear and distinguishable" test:
Information will not be considered to be provided in a clear and distinguishable manner where it can be easily overlooked or missed by natural persons under normal exposure or interaction conditions (e.g. only included as part of a manual or hidden under layers of menu options on an online interface, part of terms of use that are often not read by users).
Translated into things you can check
| Pattern | Verdict | Source |
|---|---|---|
| Disclosure only in T&Cs or the privacy policy | Insufficient | (38), (142) |
| Bot named "Ava", "Max", "Assistant" and nothing else | Insufficient | (38) |
| Behind an "i" icon, tooltip, or collapsed menu | Insufficient | (142) |
| Metadata or watermark only | Insufficient for 50(1) | (38) |
| Footer notice: "Services on this website use AI" | Insufficient | (38) |
| "This system uses LLMs" | Insufficient | (38) |
| "Powered by Intercom" | Names the vendor, not the technology | (38) by analogy |
| A tone or chime marking AI turns, with nothing spoken or written | Insufficient alone | (37) |
| Plain label in the chat, first turn, near the input | The recommended shape | (37) |
How often is enough?
Less often than you might fear. Paragraph (40): a single prominent notification before the first interaction
is likely to suffice in most instances
and paragraph (143) sets the floor at once at the start of an interactive session. Persistent badges are recommended, not required. The guidelines actively warn against overdoing it — paragraph (39) cautions about "habituation effects (so-called 'banner blindness')".
Riskier contexts change this. Paragraph (40) lists them: vulnerable users, sustained or immersive interactions, financial, insurance, legal, health advice and complaints handling, AI companions, and AI agents. In those, expect to need reminders.
One more requirement, easily missed
Also in paragraph (40):
Providers must also design and develop the AI system to ensure disclosure in all situations where the AI system is being asked questions relating to its nature or to the origin of the interaction…
No snippet satisfies this — it lives in the model's instructions. We wrote about it separately →