EU AI Act Chatbot Transparency: What Article 50 Requires from 2 August 2026
From 2 August 2026, Article 50 of the EU AI Act requires telling users they're talking to an AI. What the rule says, who is liable, and the Article 99 fines.
Yes. From 2 August 2026, Article 50(1) of the EU AI Act — Regulation (EU) 2024/1689 — requires any AI system built to interact directly with people to be designed so those people are told they are dealing with an AI, unless that is obvious. The duty falls on the provider that builds the system, but the business that deploys a chatbot on its own site has to make sure the notice appears. Penalties sit in Article 99, not Article 50.
What exactly changes on 2 August 2026?
The EU AI Act — formally Regulation (EU) 2024/1689 — was published in the Official Journal of the EU on 12 July 2024 and entered into force on 1 August 2024. Its obligations switch on in waves. The general application date, set by Article 113 (European Commission text, 12 July 2024), is 2 August 2026, with only three narrow blocks carved out for different dates: Chapters I and II from 2 February 2025; parts of the high-risk and governance machinery from 2 August 2025; and the high-risk classification rule in Article 6(1) from 2 August 2027.
The chatbot rule lives in Article 50, inside Chapter IV on transparency obligations. Chapter IV appears in none of those exceptions, so it applies on the general date: 2 August 2026. In round terms that is two years after entry into force, though the binding date is the exact one the text names — 2 August 2026 — not a loose “24 months later”.
What does Article 50(1) actually require — and when is the notice unnecessary?
The core duty is disclosure at first contact. The provider must build the system so the person knows an AI is on the other side, as Article 50(1) (European Commission text, 12 July 2024) puts it:
“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned 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, taking into account the circumstances and the context of use.” — Regulation (EU) 2024/1689, Article 50(1)
Two things deserve attention. First, the trigger is broad: any system “intended to interact directly with natural persons”, which is exactly what a website chatbot does. Second, the exception — “unless this is obvious” — is narrower than it looks. It is judged from the standpoint of a reasonably well-informed, observant and circumspect person, and Recital 132 (European Commission) adds that where the system is also meant to interact with people who are vulnerable by age or disability, that has to be taken into account. When in doubt, disclose.
How and when must the notice appear?
Clearly, distinguishably, and up front. Article 50(5) requires the information to be provided “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure”, and to “conform to the applicable accessibility requirements”.
For a chatbot that means a visible line at the start of the conversation — not a clause buried in a privacy policy, and not a disclosure that only appears after the user has typed three messages. “At the latest at the time of the first interaction” is the deadline, and the accessibility requirement means the notice has to work for screen readers and assistive technology too.
Does Article 50 stop at chatbots?
No — the chatbot notice is only paragraph 1 of four. Article 50 covers four distinct situations, and a business can fall under more than one:
| Article | Situation | Who the duty names |
|---|---|---|
| 50(1) | Systems that interact directly with people (e.g. a website chatbot) | Provider |
| 50(2) | Marking AI-generated synthetic content — audio, image, video, text — in a machine-readable format | Provider, incl. general-purpose AI |
| 50(3) | Emotion-recognition and biometric-categorisation systems | Deployer |
| 50(4) | Deep fakes and AI-generated text on matters of public interest | Deployer |
Paragraph 2 is the one to watch if you publish AI-generated media: per Article 50(2), providers of systems that generate synthetic audio, image, video or text must ensure the output is “marked in a machine-readable format and detectable as artificially generated or manipulated”. Paragraphs 3 and 4 shift the duty onto the deployer — the business actually using the system — for emotion recognition, biometric categorisation, and deep fakes.
Who is on the hook — the provider or the business deploying a third-party chatbot?
This is where most SMEs get confused. Article 50(1) names the provider — the party that builds and develops the system. A small business that drops a third-party chatbot onto its website is usually a deployer, not a provider, and paragraph 1 does not impose a literal notice duty on the deployer the way paragraphs 3 and 4 do.
That is not a loophole. In practice the user only sees the notice if the deployment shows it, so the sensible reading is: choose a provider whose product already discloses, require that disclosure in the contract, and confirm the notice actually renders on your pages. If you have wired a chatbot into your lead flow — the kind of automation we describe in WhatsApp automation for small business — you are the one who has to see the “you’re talking to an AI” line appear. Treat it as your responsibility even where the letter of paragraph 1 points at the vendor. (This is a practical interpretation, not a quote from the text.)
What is the fine — and why is it in Article 99, not Article 50?
Article 50 sets the duty; it says nothing about money. The penalties live in Article 99. Breaching the transparency obligations of Article 50 is caught by Article 99(4), which lists Article 50 by name in point (g):
“Non-compliance with any of the following provisions … shall be subject to administrative fines of up to EUR 15 000 000 or, if the offender is an undertaking, up to 3 % of its total worldwide annual turnover for the preceding financial year, whichever is higher: … (g) transparency obligations for providers and deployers pursuant to Article 50.” — Regulation (EU) 2024/1689, Article 99(4)
Watch out for a very common error: the headline “EUR 35 million or 7%” figure does not apply here. That higher tier is Article 99(3), and it is reserved for breaching the prohibited practices in Article 5 — not transparency. Two tiers, two very different numbers:
| Breach | Article | Maximum fine |
|---|---|---|
| Prohibited AI practices (Article 5) | 99(3) | EUR 35 million or 7% of worldwide annual turnover, whichever is higher |
| Transparency duties (Article 50) | 99(4) | EUR 15 million or 3% of worldwide annual turnover, whichever is higher |
There is a further nuance for smaller companies. Under Article 99(6), for SMEs including start-ups each fine is capped at the lower of the percentage or the fixed euro amount — not the higher. So the “whichever is higher” default flips to “whichever is lower” for a small business.
Does the Digital Omnibus push this deadline back?
Not for the chatbot notice. The EU’s “Digital Omnibus” simplification package is mainly about deferring high-risk AI obligations, and the Article 50(1) interaction notice is not one of them. Crucially, no modifying act changing these dates had been published in the Official Journal at the time of writing — and until an act is published there, its dates are not in force. So the operating assumption stands: the chatbot notice applies from 2 August 2026. Before you rely on any reported delay, check the consolidated text of Regulation (EU) 2024/1689 on EUR-Lex for a modifying act; if it is not there, it is not law yet.
What should an SME set up before 2 August 2026?
A short, low-cost checklist that mostly costs attention:
- Add a first-contact notice. If you run a chatbot, make sure it opens with a clear line that the user is talking to an AI — visible, not hidden in the privacy policy, shown at the first interaction (Article 50(5)).
- Check accessibility. Confirm the notice is exposed to screen readers and meets your usual accessibility requirements.
- Pin down the vendor. If the bot is a third-party product, require in the contract that it discloses its AI nature, and test that the notice actually renders on your pages.
- Label AI-generated content. If you publish AI-made images, audio, video or text, plan for the Article 50(2) machine-readable marking.
- Record the source. Keep a note of which article each decision rests on (50(1), 50(5), 99(4)) so a later audit — or a customer complaint — can be answered from the text, not from memory.
The same AI tools that help customers find you in local AI search now come with a disclosure duty when you put one on your own site, and the fix is small if you handle it before the deadline. Wiring a compliant chatbot into your lead flow — notice included — is exactly the kind of build we handle in our process automation service: the disclosure ships with the flow, not as an afterthought.
This article is general information, not legal advice. It does not replace reading the official text on EUR-Lex or consulting a qualified lawyer. For any specific case, check the consolidated version of Regulation (EU) 2024/1689 and any modifying act published in the Official Journal of the EU.
Frequently asked questions
Do I have to tell users my chatbot is an AI?
Yes. Article 50(1) of Regulation (EU) 2024/1689 (the EU AI Act) requires AI systems built to interact directly with people to be designed so those people know they are dealing with an AI. The only exception is where that is obvious to a reasonably well-informed, observant and circumspect person given the context and circumstances of use. The literal duty falls on the provider that builds the system; in practice, a business deploying a chatbot on its site should make sure the notice is actually shown to the user.
When does this obligation start?
From 2 August 2026. Article 113 sets that as the Regulation's general application date, and Article 50 (Chapter IV) is not among the exceptions with different dates. The Regulation was published in the Official Journal of the EU on 12 July 2024 and entered into force on 1 August 2024, so the chatbot notice applies 24 months later.
How and when must I show the notice?
Article 50(5) requires the information to be given in a clear and distinguishable manner, at the latest at the time of the first interaction, and in line with the applicable accessibility requirements. For a chatbot that means a visible notice at the start of the conversation, not a line hidden in the privacy policy or shown only after several messages.
What if it is obvious my chatbot is a bot?
Article 50(1) exempts cases where interacting with an AI is obvious to a reasonably well-informed, observant and circumspect person. It is a narrow exception, so when in doubt, disclose anyway. Recital 132 also asks you to consider vulnerable groups by age or disability where the system is meant to interact with them.
What fine could I face for not disclosing?
The penalty is in Article 99, not Article 50. Breaching the transparency duties of Article 50 falls under Article 99(4): up to EUR 15 million or, if the offender is a company, up to 3% of total worldwide annual turnover for the preceding year, whichever is higher. The higher EUR 35 million / 7% tier in Article 99(3) is reserved for the prohibited practices of Article 5, not transparency. For SMEs and start-ups, Article 99(6) caps the fine at the lower of the two amounts.
Does the Digital Omnibus delay the chatbot notice?
Not for the Article 50(1) interaction notice, which the reported simplification package does not target. As long as no modifying act has been published in the Official Journal, the dates in the Regulation stand and the chatbot notice applies from 2 August 2026. Always check the consolidated text on EUR-Lex before assuming any change.