EU AI Act and Chatbots: What You Must Do Before August 2, 2026

Most AI Act coverage talks about high-risk systems and model providers. For the vast majority of businesses the part that actually bites is much smaller and much closer: Article 50, the transparency obligations. They apply to every company that lets an AI system interact directly with people – which is exactly what a website chatbot or an AI phone agent does. The rules become applicable on August 2, 2026.

What Article 50 requires

The core duty is simple to state: people must be informed that they are interacting with an AI system. The EU AI Act names the affected categories explicitly – chatbots, virtual assistants and automated phone systems. Two details decide whether you are compliant:

• The disclosure must live in the interaction itself. According to the European Commission's draft guidelines, a statement buried in your terms and conditions does not satisfy the duty – and neither does calling the bot a vague "assistant". The user has to be able to perceive it where the conversation happens.
• It applies to voice, too. An AI phone agent has to make clear at the start of the call that the caller is talking to an AI, not a human.

Who is responsible – you or your chatbot vendor?

Both, in different roles. Providers must design their systems so users are informed. But as the company deploying the chatbot toward your customers, you stay responsible for how it operates on your website and your phone line – buying from a vendor does not outsource the obligation. That makes vendor selection a compliance decision: ask which foundation model runs underneath, whether the provider meets its own AI Act duties, and get a data processing agreement that reflects it.

The stakes are real: transparency violations can be fined with up to 15 million euros or 3 percent of global annual turnover, whichever is higher.

The 6-point checklist before August 2

1. Inventory your AI touchpoints. Website chat, WhatsApp bot, AI phone number, voice assistant – anywhere an AI talks to customers directly.
2. Check the disclosure in the interface. Open your own chat widget as a visitor: does it say, visibly, that answers come from an AI? A brand name or "assistant" label is not enough.
3. Check the phone channel. Call your own AI number: does the greeting identify it as an AI assistant before the conversation starts?
4. Collect vendor evidence. Which model powers the bot, where is data processed, does your data processing agreement cover the AI Act? Keep the answers on file.
5. Define the human handover. When a human agent takes over the chat, the AI label should disappear – and customers should be able to reach a human for cases the bot cannot handle. Not an Article 50 duty in itself, but it keeps the disclosure truthful in both directions.
6. Document it. A dated screenshot of the disclosure in your widget and a recording note for the phone greeting are cheap insurance if anyone ever asks.

How WebChatAgent handles this for you

We shipped Article 50 compliance as a default, not a setting:

• Web widget: every AI reply carries an "AI Agent" label right under the message, automatically, in all 13 widget languages. When a human agent takes over, the label disappears, because then a human is writing.
• AI phone agent: the default greeting introduces itself as an AI assistant, and if a caller interrupts the greeting, the assistant discloses it in its next reply.
• EU processing: with the default models, processing runs in the EU (Google Vertex AI, Frankfurt/EU region), and a data processing agreement is included – details and model options in our GDPR overview.

If you run your bot with us, points 2 and 3 of the checklist are already done. The inventory, vendor file and documentation remain yours – they take an afternoon.

FAQ

Does this apply to small businesses?

Yes. The transparency duty is about protecting the people talking to your bot, not about your company size. If an AI interacts directly with your customers, the disclosure duty applies.

Is naming the bot "Assistant" enough?

No. The Commission's draft guidelines explicitly say a vague reference to an "assistant" does not satisfy the duty. Say it plainly: the user is talking to an AI.

Does it cover AI phone systems?

Yes – automated phone systems are named alongside chatbots and virtual assistants. The caller must learn at the start of the call that they are speaking with an AI.

What about existing chats or old transcripts?

The duty concerns the interaction as it happens. Make sure every conversation from August 2, 2026 onward carries the disclosure; you do not need to retrofit past transcripts.

This article is general information, not legal advice. For your specific setup, especially if you operate high-risk systems, talk to a lawyer.