Your chatbot now has to say it's an AI: what changed on 2 August
Since 2 August 2026 you must tell people when they're talking to an AI system. Who it affects, what exactly you need to say, and what happens if you don't.
On 2 August 2026 the transparency obligations of the European AI Act came into force. The most concrete one, and the one affecting the most businesses, is this: if a person is talking to an artificial intelligence system, they have the right to know it.
This isn't a recommendation or a best practice. It's Article 50, it has been enforceable for a few weeks now, and it applies to any company using a chatbot, a virtual assistant or a voice agent to serve its customers. It doesn't matter whether you built it or bought it.
Who it actually affects
There's a widespread misunderstanding here: many people assume the AI Act is about companies that build artificial intelligence. It is about that too, but Article 50 reaches whoever uses it.
If your business has any of these, it applies to you:
- A chatbot on your website answering customer questions.
- A WhatsApp assistant that replies, books appointments or resolves queries.
- A voice agent that picks up the phone.
- Any system generating automated AI replies inside a conversation.
And you're not off the hook because you hired an agency. The obligation falls on whoever deploys the system to the public — that is, on your business.
What you have to do, specifically
The law requires the notice to be clear and to arrive, at the latest, at the moment of first interaction. In practice: the person must know they're talking to a machine before telling it anything, not in the small print and not three messages later.
In a web chat, it's enough for the assistant to introduce itself for what it is in its first message. Something this simple already complies:
Hi, I'm [business]'s virtual assistant. I'm an AI, so if you'd rather speak to a person just say so and I'll pass you to the team.
Notice the two halves. The first declares what it is. The second offers a way out to a human — not mandatory in every case, but it prevents that feeling of being trapped talking to a machine, and that feeling is what makes people hang up.
Same thing for a voice agent, said in the first seconds of the call. And on WhatsApp, in the bot's first message.
What does NOT comply
- Putting it only in the privacy policy or legal notice. Nobody reads them, and the law wants the notice inside the interaction itself.
- A robot-sounding name and nothing else. Your bot being called «Nova» informs no one: people are called that too.
- Saying it at the end of the conversation, once the person has explained their problem and handed over their details.
- A notice hidden in an icon or in pale grey text nobody sees.
What happens if you don't
For breaches of the transparency obligations, the AI Act sets a ceiling of 15 million euros or 3% of worldwide annual turnover, whichever is higher.
Before your stomach drops: that's the theoretical maximum and it's aimed at large platforms. No small business is getting fined 15 million over an unlabelled chatbot. But the framework exists, in Spain it's supervised by AESIA (the Spanish AI Supervision Agency), and since August it has inspection powers.
And there's a cost more likely than the fine: distrust. The day a customer realises halfway through that they've spent ten minutes explaining their problem to a machine posing as a person, they don't come back. That happens long before any inspection does.
The part almost nobody mentions: declaring it sells better
There's a belief that if you say it's a bot, people won't use it. Our experience says the opposite.
When the assistant introduces itself as an AI from the start, the conversation changes: people ask more directly, without preamble, and they accept an instant answer at eleven at night far more readily. What bothers people isn't talking to a machine — it's finding out they were fooled.
Every assistant we build identifies itself as an AI in its first sentence and offers a handover to a person when needed. We don't do it out of fear of a fine: we do it because it works better.
What to check this week
If you have a conversational system running, this check takes five minutes:
- Open your own chat as if you were a customer. Does the first message say it's an AI?
- Call your own number if you have a voice agent. Does it say so in the first seconds?
- Message your WhatsApp bot from a number that isn't yours. Same check.
- Confirm there's a way to ask for a human, and that the route actually works.
If any answer is no, it's a quick fix: one sentence in the assistant's configuration. Complying isn't the expensive part — finding out late is.
Want to know what you could automate?
Tell us what eats your hours and we will send back a concrete plan with a closed price in under 48 hours.