EU AI Act's Chatbot Disclosure Rule, Explained
If your business runs (or is considering) a chatbot on your website, there's a compliance deadline worth knowing about even if you've never heard of the EU AI Act: Article 50's transparency obligations become enforceable on August 2, 2026. In plain terms, any chatbot or virtual assistant interacting with the public has to clearly disclose that people are talking to AI, and it has to say so from the start of the interaction, not buried in a footer or a terms page.
Who this applies to
This is the detail that catches people out: the obligation sits with the business deploying the chatbot, not just the company that built the underlying AI model. If you run a chatbot on your site (whether you built it yourself, had someone build it for you, or bought an off-the-shelf tool), you're the deployer, and you're the one responsible for the disclosure being there.
It doesn't matter whether the bot is a simple FAQ widget or something closer to a full conversational agent. If it's AI, interacting with a natural person, and that person could reasonably think they're talking to a human, the disclosure requirement applies.
What "disclosure" means
The requirement isn't complex, but it is specific:
- The disclosure has to happen at the start of the interaction: not after several exchanges, not only when asked.
- It has to be clear, not hedged or vague. "This chat may sometimes use automated assistance" doesn't meet the bar. "You're chatting with an AI assistant" does.
- It has to be visible, not technically present but practically invisible. A disclosure in 6-point grey text isn't going to hold up.
That's a low technical bar to clear. A single line at the top of the chat window, present every time a new conversation starts, satisfies it. The harder part isn't building the disclosure. It's remembering it's not optional, and that "we'll add it later" isn't a safe place to leave it once August arrives.
Why this isn't a small thing to skip
Fines for AI Act violations aren't scaled to the size of the business breaking the rule. They can reach €15 million or 3% of global annual turnover, whichever is higher. A large platform can shrug off a fine like that. A solo practice can't — and doesn't have to risk it. The fix costs nothing and takes minutes.
What we do about it
Every chatbot we build ships with the AI disclosure line active by default. It's not a setting a client has to remember to switch on, and it's not a line item we bill separately. It's baked into the build the same way error handling or basic accessibility would be: a default, not an add-on.
Practically, this shows up as a short, clear statement the moment a chat window opens, something like "You're chatting with [Business]'s AI assistant", before any other message, on every new conversation, regardless of the channel the chatbot runs on.
If you already have a chatbot
If you're running a chatbot today that doesn't disclose it's AI from the outset, the fix is almost always a small one: add the line, make sure it fires at the true start of every conversation, and make sure it's not something a future update could silently remove. Not sure if your current setup meets the bar? Check now — it takes minutes, and August isn't optional.
For more on how we think about disclosure and honesty in what we build generally, see why Fluxlane exists.