Since August, your chatbot has to admit it's a bot

I went through our own site while writing this and found a problem. Dante, the chat assistant in the bottom corner, opened every conversation with "Hi, I'm Dante, Mindaptive's assistant." Assistant. Not AI assistant. We build these things for a living and still missed it, so I'm guessing plenty of other businesses did too. (It's fixed now.)

Here's why it matters. On 2 August 2026 the transparency rules in Article 50 of the EU AI Act started to apply. The short version: if an AI system talks to people, they have to be told they're dealing with an AI. The exception is when it's obvious from context, and I wouldn't bet a fine on "obvious."

Wait, wasn't the AI Act delayed?

Partly, and that's where a lot of the confusion comes from. In June the EU signed off on the so-called Digital Omnibus, a package that pushed back the rules for high-risk AI systems. Those were supposed to start this August and now start in December 2027 (August 2028 for AI built into regulated products). A few headlines made it sound like the whole Act was on hold. It isn't.

The transparency duties weren't part of the delay. The Omnibus did give systems already on the market until 2 December to add machine-readable marking to AI-generated content, but a bot chatting with your customers without saying what it is gets no grace period.

Who this actually hits

If you run a small business you probably aren't training AI models. You might well be running a chat widget that answers questions and books calls, though, or a voice agent that picks up the phone when you can't, or a WhatsApp bot replying to leads at 11pm. All of those talk to people.

Strictly speaking, Article 50(1) puts the duty on the provider, meaning whoever built the system. In practice, if an agency built it and you switched it on, your name is on the website and you're the one customers will be annoyed with. I'd treat it as yours.

Fines for breaking the transparency rules can reach €15 million or 3% of worldwide turnover. For SMEs it's whichever of the two is lower. Nobody is going to fine a dental practice in Split €15 million over a chat bubble. Still, "we probably won't get fined" is a pretty weak reason to mislead the people you're hoping will become customers.

What we changed, and what I'd tell a client

The fix is boring. For chat, say it in the first message. "Hi, I'm Dante, Mindaptive's AI assistant" does the job. Don't hide it in your terms of service or a tooltip nobody hovers over.

Voice is a little harder because people hate long intros on the phone. We keep it to one sentence at the start of the call, something like "Hi, you've reached the clinic, I'm the AI receptionist, how can I help?" It takes about two seconds.

While you're in there, make sure people can reach a human when they want one, and that the bot never claims to be a person if someone asks it straight out. I'd also skip the trend of giving the bot a real employee's name and photo. It feels friendlier right up until a customer figures it out.

If you don't know what your own chatbot says when it opens, go and check. It takes thirty seconds, and ours was wrong.