Yes. If your AI sales agent talks to anyone in the EU, you are required to tell them they are talking to an AI, and that rule has been enforceable since 2 August 2026. If it runs on WhatsApp, Meta's messaging policy adds a second requirement: the customer must always be able to reach a human. Neither rule is hard to satisfy, and neither has to hurt your funnel, but both have to be designed in rather than bolted on.
What the law actually says
The relevant text is Article 50 of the EU AI Act, Regulation (EU) 2024/1689, which you can read in full on EUR-Lex. The transparency obligations are short and unusually plain for EU legislation:
- Article 50(1): providers of AI systems intended to interact directly with natural persons must design them so that those people are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person.
- Article 50(2): providers of systems that generate synthetic text, audio, image or video must mark the output as artificially generated in a machine-readable format, where technically feasible.
- Article 50(4): deployers who publish AI-generated text on matters of public interest must disclose it, unless a human editorially reviewed it and takes responsibility.
For a sales or support agent, 50(1) is the one that bites. The European Commission's own AI Act overview confirms the phased timeline: prohibited practices in February 2025, general-purpose model duties in August 2025, and most remaining obligations including Article 50 from 2 August 2026. The penalty ceiling for breaching these transparency duties, under Article 99, is the higher of 15 million euros or 3 percent of worldwide annual turnover.
Two points founders often miss:
- Scope is about the user, not you. The Act applies to systems used by people located in the EU. A Dubai clinic with an Arabic-speaking WhatsApp agent still triggers it the moment a patient messages from Paris.
- "Obvious" is a weak defence. A chat bubble that says "Hi, how can I help?" in a natural tone and answers in under a second is not obviously a machine to an ordinary customer. The regulator's test is the customer's perception, not yours.
WhatsApp has its own rule, and Meta enforces it faster than any regulator
If your agent runs on the WhatsApp Business Platform, read the WhatsApp Business Messaging Policy. It requires that businesses using automation clearly disclose that the person is interacting with an automated experience, and that they provide a clear way to request transfer to a human agent. Meta can throttle or shut a business number for violations, and appeals move at Meta's pace, not yours.
This matters more than the fine in practice. A regulator might take a year to notice you. A number ban takes your revenue channel offline on a Tuesday afternoon. If you are pricing a deployment, the escalation path is not optional scope; we covered how to build it in AI agent escalation paths, and it should already be in the estimate you got for what an AI WhatsApp sales agent costs.
Outside the EU
Other jurisdictions are converging on the same shape, and the one-line disclosure shown in the next section already covers most of them. California has required since 2019 that a bot used to sell goods or services identify itself, which a first message containing the words "AI assistant" does in its opening sentence. Several Gulf regulators publish AI ethics principles that name transparency toward end users, and public-sector and healthcare tenders in Saudi Arabia and the UAE increasingly ask for it explicitly. Go back to the Dubai clinic: one patient messaging from Paris pulls it under Article 50, a hospital tender at home asks for the same disclosure in writing, and a customer with a California number is covered by the 2019 bot law. Three rulebooks, one first message. Designing to the EU standard once is cheaper than tracking each market separately.
The design question: how to disclose without killing conversion
The fear is real: you built a sales agent because it converts, and now you are being told to open every conversation with a disclaimer. Here is what actually works.
Say it once, early, in plain words
One line at the start of the conversation is enough for Article 50(1). Keep it short and human:
Hi, I'm Layla, the AI assistant for Northwind Clinic. I can book appointments and answer questions. Type "human" any time to reach our team.
That single message satisfies both the AI Act and WhatsApp. It also does something useful for you: it sets expectations, so when the agent asks a qualifying question the customer does not feel interrogated by a stranger.
Do not fake a human name without the label
Giving the agent a name is fine. Giving it a name, a profile photo of a real-looking person, and no disclosure is exactly the pattern the law was written to stop. If you want a persona, pair it with the word "AI" or "assistant" in the first message and in the WhatsApp business profile description.
Make the handoff explicit, both ways
When the conversation moves to a person, say so: "Passing you to Omar from our team." When a human hands back to the agent after hours, say that too. Customers tolerate automation; they do not tolerate being unsure who they are talking to.
Keep the machine-readable marker in mind
If your agent sends generated documents, quotes or images, Article 50(2) expects a machine-readable marker such as metadata or a content credential. Most teams can satisfy this by keeping generated PDFs and images on a template pipeline that stamps provenance metadata. Ask your vendor how they handle it before you sign.
What we learned from our own outreach
We run our own outbound engine. It scrapes a prospect's website, uses a local model to pull out concrete facts about the company, and drafts one tailored email per prospect in a single call. The part relevant here is where the automation stops: a person reviews and sends, and a person answers every reply.
That boundary turned out to be commercially useful, not just compliant. We tested the call to action and found that asking for a simple "reply YES" consistently beat linking to a calendar. The reason is friction, but the effect on disclosure is the same: the moment someone replies, they are talking to a human, and we can say so honestly. We wrote more about the CTA finding in AI sales agent CTA design. If your funnel has a natural point where a human takes over, put the handoff there and be explicit about it. You get better replies and a cleaner compliance story at the same time.
A short checklist before you go live
- First message discloses AI in the user's language, and it renders in the WhatsApp preview.
- A human escape hatch exists, tested on a real phone, and it works outside business hours (even if the answer is "a person will reply at 9am").
- Business profile description on WhatsApp mentions automated assistance.
- Handoff messages are explicit in both directions.
- Generated files carry provenance metadata, or you have a written reason why it is not technically feasible.
- Logs record which turns were AI and which were human, so you can prove compliance if asked.
- Contract with your vendor names who is the "provider" and who is the "deployer" under the Act, because the obligations differ.
The real risk is not the fine
The 15 million euro ceiling in Article 99 is the number everyone quotes, but it is not what should worry you. An AI agent that pretends to be a person, gets caught, and gets screenshotted is a brand problem before it is a legal one, and a WhatsApp number ban lands months before any regulator writes to you. Everything the rules ask for is already in the Layla message above: who the customer is dealing with, what it can do, and the one word to type to reach a person. Our own outreach passes the same test not because we wrote a policy but because the human boundary is real; the moment a prospect replies YES, a person is reading. Build the disclosure, the escape hatch and the AI-versus-human logs into the first sprint and you will never think about Article 50 again.
If you are scoping a WhatsApp or web sales agent for a market that includes the EU or the Gulf and want the disclosure, escalation and logging handled properly from day one, let's talk.
Originally published on the Pykero blog.
Top comments (0)