The bot has to say it is a bot
Somebody set up the chat widget on your site a while ago. They gave it a name, because the field asked for one, and a photograph, because that field was there too. Maybe it is called Aria. Maybe it has your logo. It answers questions about opening hours at half past eleven at night, and most visitors never wonder what is on the other end.
Nobody in that story set out to deceive anyone. The name field said "name" and somebody typed a name.
There is now a piece of European law about exactly this moment, in application since 2 August 2026. The rule itself is short. What took me longer to get over is that the products people build these assistants with mostly have no setting for it, and the two vendors that do address it in the same screen give opposite advice.
I am not a lawyer, this is not advice about your obligations, and whether any of it reaches your business depends on where you and your customers are. What follows is the legislative text, the dates, and what seventeen products actually document.
What the rule says
Article 50(1) of the EU AI Act, formally Regulation (EU) 2024/1689, quoted whole:
"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. This obligation shall not apply to AI systems authorised by law to detect, prevent, investigate or prosecute criminal offences, subject to appropriate safeguards for the rights and freedoms of third parties, unless those systems are available for the public to report a criminal offence."
Regulation (EU) 2024/1689, Article 50(1)
And the timing, from paragraph 5 of the same article:
"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."
Two things about that text before anything else.
It binds providers. Paragraph 1 says "Providers shall ensure". A business that buys a chat product and switches it on is normally a deployer, and the deployer duties in this article are paragraphs 3 and 4, about emotion recognition and deep fakes. So the sentence above is aimed at the company that built your widget. Keep that distinction; the vendors' contracts turn on it.
The exemption is a judgement call. The duty falls away where the machine nature "is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect". A widget named Aria with a human face is precisely the case where nobody can say in advance how that test comes out.
The date, because the search results have it wrong
Search for this and you will be told the transparency rules landed in July 2025. They did not, and the text settles it in two steps.
Article 50 lives in Chapter IV of the Regulation, headed "TRANSPARENCY OBLIGATIONS FOR PROVIDERS AND DEPLOYERS OF CERTAIN AI SYSTEMS". It is the only article in that chapter.
Article 113 then sets out when things apply:
"It shall apply from 2 August 2026.
However:
(a) Chapters I and II shall apply from 2 February 2025, with the exception of Article 5(1), first subparagraph, points (ba) and (bb), and Article 5(1a) and (1b) which shall apply from 2 December 2026;
(b) Chapter III Section 4, Chapter V, Chapter VII and Chapter XII and Article 78 shall apply from 2 August 2025, with the exception of Article 101;
(c) Chapter III, Sections 1, 2, and 3, with the exception of Article 6(5), shall apply from:
(i) 2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2) and Annex III; and
(ii) 2 August 2028 as regards AI systems classified as high-risk pursuant to Article 6(1) and Annex I;
(d) Articles 102 to 110 shall apply from 27 July 2026."
Regulation (EU) 2024/1689, Article 113, consolidated version of 27 July 2026
Chapter IV appears in none of those exceptions, so the general rule governs and Article 50 applies from 2 August 2026. The European Commission says the same thing in its own words: "The transparency rules of the AI Act will come into effect in August 2026."
One more distinction worth holding, because the two dates get swapped constantly. Entry into force was 1 August 2024. Application of this article was 2 August 2026. A regulation that is in force is not yet a regulation you are being measured against.
A note on which text to read, since this caught me out. The Regulation has been amended, by Regulation (EU) 2026/1744 of 8 July 2026, and the consolidated version dated 27 July 2026 rewrites subparagraph (c) of Article 113 and adds a subparagraph (d) that did not exist before. Article 50's first six paragraphs are untouched; only paragraph 7, about codes of practice, changed. Anyone quoting the original 2024 text on dates is quoting something superseded.
What the products actually do
The law was the short half. I read the documentation of seventeen chat and assistant products and platforms on 8 September 2026, looking for one thing: does the product tell the visitor it is a machine, and if so, who decided?
| Product | Disclosure feature? | Default | Does the naming or persona page mention it? |
|---|---|---|---|
| Intercom | Yes, a named feature | Messenger label off unless the workspace used Fin in Messenger by 10 July 2025 | No |
| Zendesk | One advisory sentence | Advice, not a control | On the name page, not the persona page |
| Tidio | Advice only, on a legacy page | Advice, not a control | No |
| Chatbase | Email disclaimer only | Off, opt-in | No |
| Salesforce | Required, but of the customer by contract | Policy, not a product default | Not confirmed |
| Freshworks | Nothing, and advises against machine-sounding names | Not applicable | No |
| HubSpot | Nothing | Not applicable | No |
| Microsoft Copilot Studio | Nothing about identity | Not applicable | Not addressed |
| Crisp | Nothing | Not applicable | No |
| Landbot | Nothing | Not applicable | No |
| ManyChat, Voiceflow, Botpress | Nothing findable, except one negative duty at Voiceflow | Not applicable | No at Voiceflow; not checked for the other two |
| Meta, Messenger and Instagram | Required, conditional on applicable law | Not applicable | Not applicable |
| Meta, WhatsApp Business | Nothing about disclosure | Not applicable | Not applicable |
| Telegram | Nothing | Not applicable | Not applicable |
| Discord | Nothing | Not applicable | Not applicable |
That table describes documentation pages read on one day. Products change, and several of these are actively being rewritten.
The one product that ships the feature, switched off
Intercom is the only vendor in the set with a documented feature called AI Agent Disclosure. Its help page opens by explaining why it exists, and then says this:
"This toggle is on by default for all customers using Fin AI Agent in the Messenger up until the 10 July 2025."
"This toggle is off by default for all other customers."
Intercom, AI Agent Disclosure, dated 30 June 2026
If your workspace adopted Fin after 10 July 2025, that paragraph is about you. The label that tells your visitor they are talking to software exists, is documented, and is off unless somebody went and turned it on. The email footer setting is the other way round, on by default, and produces the line "This answer was composed by [AI Agent name], [Workspace name]'s AI Agent."
The same page then puts the duty where Article 50 does not, on the customer:
"Note: Under our Additional Product Terms for AI Product (Section 5, Transparency) our customers are responsible for ensuring that their end users are informed that they are interacting with an AI system. Therefore, we suggest that you always employ one or both of the AI Agent label and introductory message containing a reference to Fin being an AI Agent. We are not responsible for any consequences arising from disabling both of these features."
So the statutory duty in paragraph 1 sits on the provider, and the provider's contract puts the practical duty back on you. Both directions arrive at the same screen, which is the one nobody has opened since setup.
Two vendors, one field, opposite instructions
This is the finding I did not expect. The name your assistant shows to customers is a single text box in every one of these products. Two vendors give guidance in that exact box, and it points opposite ways.
Zendesk, in step 5 of updating an assistant's name:
"In Name, update the AI agent's name. This name appears at the top of the Web Widget in messaging conversations and as the sender in emails. Choose a name that makes it clear customers are not talking to a human."
Zendesk, configuring an AI agent's name and avatar
Freshworks, in the same field:
"Choose a name that fits your brand - for example, 'Aria', 'Max', or 'FreshCart Support'. Avoid generic names like 'Bot' or 'AI'; named agents have higher customer engagement and lower abandonment rates."
Freshworks, configure AI Agent identity and instructions
One tells you to make the machine obvious. The other supplies a commercial argument for the opposite, with engagement and abandonment as the reasons. A business that switched helpdesks in 2025 has followed both, in order, without either being wrong at the time.
And the Zendesk page carries a banner: the functionality it describes "moved into legacy status on July 31, 2025" and "On December 10, 2026, this functionality will no longer be available." The single clearest instruction anybody gives about this, in the field where it would actually be read, is on a page scheduled for removal.
Same company, two channels, two answers
Meta's Messenger platform policy sets out a disclosure duty in detail, including suggested wording:
"When required by applicable law, automated chat experiences must disclose that a person is interacting with an automated service: at the beginning of any conversation or message thread, after a significant lapse of time, or when a chat moves from human interaction to automated experience."
Meta, Messenger Platform policy
It names California and Germany as markets deserving special attention. It does not mention the AI Act. And the duty is conditional on law rather than imposed by Meta, which means the policy hands the question straight back to you.
The WhatsApp Business messaging policy, from the same company, requires something different and never requires disclosure at all:
"You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths."
WhatsApp Business messaging policy
The exit to a human is mandatory; the heads-up that you are not talking to one goes unmentioned. Run the same assistant on both channels and it lives under two different sets of expectations.
The line everybody still quotes has been deleted
There is a widely repeated claim that OpenAI requires developers to tell users they are interacting with an AI. The current usage policies, effective 29 October 2025, contain no such requirement. What survives is narrower and lives on a separate page about ChatGPT agent: a prohibition on "Misrepresenting or concealing the role of AI technology in tasks or interactions". That is a ban on concealing, which is not the same as a duty to announce, and it is scoped to a product rather than the API.
Worth knowing if your policy on this was written by copying a vendor's.
Which of your customer-facing tools speaks in your name, and what does each one say about itself? That is one of the questions a process audit answers in writing, alongside what those tools cost you and what they are worth.
What the exposure actually looks like
The headline number attached to Article 50 in most coverage is fifteen million euros or three per cent of worldwide turnover. That figure is real, and quoting it without the next paragraph overstates what it means for a small business.
Article 99(4) sets 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", and its point (g) covers "transparency obligations for providers and deployers pursuant to Article 50."
Article 99(6) then says:
"In the case of SMEs, including start-ups, each fine referred to in this Article shall be up to the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower."
For an SME the lower of the two figures applies. And Article 99(1) leaves the penalty regime to Member States, which "shall take into account the interests of SMEs, including start-ups, and SMCs, and their economic viability when imposing penalties". The figure is a ceiling on national law rather than a price list.
Set the fine aside, then. What makes this rule worth a small business's attention is that the disclosure question is decided, today, by a default in a settings screen, and defaults are the part of a system nobody remembers choosing.
A different structure, for contrast
California has had a bot statute since 2019, and comparing it with Article 50 shows how differently the same worry can be legislated. California Business and Professions Code section 17941:
"(a) It shall be unlawful for any person to use a bot to communicate or interact with another person in California online, with the intent to mislead the other person about its artificial identity for the purpose of knowingly deceiving the person about the content of the communication in order to incentivize a purchase or sale of goods or services in a commercial transaction or to influence a vote in an election. A person using a bot shall not be liable under this section if the person discloses that it is a bot."
That is a deception offence. It needs intent, it needs a purpose from a short list, and disclosure is the defence rather than the duty. Article 50(1) has no intent element and no purpose list at all: the obligation attaches to any system meant to talk to people, and it is discharged by informing.
Two jurisdictions, two shapes. Neither of them is the one your chat widget was configured under, because your chat widget was configured under whichever defaults shipped that week.
Where this leaves you
Whatever the law ends up demanding of you, the practical question is whether anybody in your business could answer three things about the assistant currently talking to your customers.
What does it call itself, and who chose that? In most small businesses the answer is an agency, a contractor, or the person who set it up on a Friday. The name is the disclosure, whether or not anyone thought of it that way.
Is the label on? For Intercom the honest answer, for any account that was not using Fin in Messenger by 10 July 2025, is no, by default, and finding out takes about a minute. For most of the others there is no label to switch on at all, which means the greeting is carrying the whole weight.
What does it say in its first message? Paragraph 5 of Article 50 puts the timing at "the latest at the time of the first interaction", and the first message is the one thing every chat product in that table lets you edit.
That is a short list rather than a compliance project, and it belongs with whose accounts your automation runs on and which automations you actually have: things a business owns without knowing it owns them, until a day comes when the answer is needed quickly.
It is also worth separating this from the older worry about assistants, which is what happens when yours says something untrue. A tribunal has already worked through who answers for that, and the answer was the business. Disclosure is a different obligation entirely: it applies when the assistant is working perfectly, and it is about what the person on the other end knows.
If you would rather have the map than assemble it in a hurry, listing every automated thing that speaks to your customers and what each one is doing on your behalf is part of the process audit. $299, three business days, and a written map you keep whether or not you hire us.
Nothing here is legal advice, and whether any of these rules touches your business depends on where it and your customers are.
Related: comparing an AI receptionist with hiring somebody covers the decision that comes before this one.