Anyone in Europe chatting with a business online has a legal right to know whether a person or a machine is answering. That right has applied since August 2, 2026, when the transparency rules in the EU’s AI Act took effect. The right extends past the chat box in the corner of a website. It also covers a customer messaging a company on WhatsApp, asking a question on Instagram or getting an automatic reply on Facebook. Any AI system used by people inside the EU is covered, even if the company behind it sits outside Europe entirely.
Before the rules, many chatbots were built to sound as human as possible, with nothing telling the person on the other end that they were talking to AI. That was a design choice. The rules put it on the wrong side of a legal line, because the disclosure has to come at the first message someone in Europe receives.
The rule reaches into content as well. AI-made images, video and audio realistic enough to pass as genuine, the material behind deepfakes, need a label showing they were made or altered by AI. Text carries the same duty when a business publishes AI-written material meant to inform the public.
Email gets a narrower test. Some AI systems write and send replies on their own, with no person reading them first. That counts as direct interaction with someone, so such a system needs the same disclosure as a chatbot. A person who asks AI to draft an email, then reads and sends it, sits outside the rule, even though the email looks identical on the receiving end. What the rule tracks is whether a human read it before it left the outbox, because someone who reads before sending is still the one doing the talking.
The duty is easier to write into law than to police. The law says nothing about how a regulator would spot a small business whose chatbot gives no disclosure. The same goes for an email system with no one checking its replies before they go out. That part of the rule was untested when the deadline passed.
