If you have a chatbot on the site, from August 2, 2026 there is a line that you must have and that can be checked from the outside in ten seconds: the user must understand, clearly, that he is talking to a machine.
Article 50 of EU Regulation 2024/1689 establishes a transparency obligation for those who use AI systems in contact with the public. This isn't a technical obligation: it's a reporting obligation. And that's why this violation is the easiest to challenge.
Where it is needed, and where it is not
It's useful when the person might not understand it on their own. Typical cases:
- the assistant who answers questions about products;
- the quote form that "converses" instead of having you fill in fields;
- assistance via WhatsApp or messaging managed by an automatic system;
- automatic responses to reviews, if posted, appear as if they were handwritten.
There's no need to declare that you use spell checker, and there's no need to put a disclaimer on every page of your site. The rule of thumb: If there is an exchange, it must be stated at the beginning of the exchange.
The two lines that solve
Legal text isn't required. Two sentences are required in the bot's first message:
> "Hi, I'm [company's] auto-assistant. I answer using artificial intelligence, and I might make mistakes. For special cases, I'll put you through to someone else."«
It contains all the useful elements: the automatic nature, the limit, the human escape route. The latter isn't required by law, but it's what saves difficult conversations—and it's usually also the only thing the customer wanted to know.
The mistakes I see most often:
- The human name without warning. A bot that introduces itself as "Giulia" and says nothing else is the case that the law is intended to avoid.
- The hidden statement. Just writing it in the privacy policy is not enough: it must be clear In the point of contact.
- The fake person during closing hours. If at 11pm "Giulia" responds in two seconds, transparency has already effectively broken.
What are the risks of those who don't do it?
Violation of transparency obligations falls within the second sanction level: up to 15 million euros or 3% of annual global turnover, with the most favorable criterion for SMEs and start-ups.
The figure is theoretical for a small business. The real point is another: this violation is visible to anyone, customer, or competitor. You don't need an inspection to notice, just visit the site.
Do it today, it's five minutes
Open your site from your phone, incognito, and act like a customer. Check the first message that appears in the chat. If it doesn't contain the word "automatic" or "artificial intelligence," you've found your job for tonight.
Then do the same thing on the channels you tend to forget: the automatic message on WhatsApp Business, the automatic replies on Messenger, the assistant inside the customer area.
In most cases, it's a matter of changing a sentence in a configuration panel. It's the best cost-benefit ratio of the entire AI Act.
If you want me to check your contact points and write you the right sentences, write to me.
Sources
- Regulation (EU) 2024/1689 (AI Act), Article 50: Transparency obligations for systems intended to interact with natural persons, applicable from 2 August 2026
- Regulation (EU) 2024/1689, Article 99: Sanctioning regime
- Logotel, AI Act, transparency obligations: what changes for companies, 2026
- Money.it, AI Act: What's changing starting August 2nd for businesses using artificial intelligence?