EU AI Act: the transparency rules now apply to your chatbot
Since 2 August 2026 you must tell customers they are talking to AI. What the rule actually requires, which deadline follows on 2 December 2026, and what to check today.
Author: Heyloha Team
Do the EU AI Act rules already apply to chatbots?
Yes. The transparency obligations in Article 50 of the EU AI Act have applied since 2 August 2026. This is no longer a future deadline to prepare for. It is law in force.
The core is simple: an AI system that interacts directly with people must be designed so that those people know they are dealing with AI. That covers chatbots, AI telephony and voice assistants. Formally, Article 50(1) puts this duty on the provider, the party that develops the system and places it on the market. Your customers experience it as your service, though, so check it yourself as well.
The obligation applies to all in-scope systems regardless of when they were placed on the market. You cannot rely on your chatbot having existed before August 2026.
The European Commission adopted guidelines on 20 July 2026 explaining how to comply.
What exactly does your chatbot have to disclose?
The disclosure must happen at the latest at the time of the first interaction. Not halfway through the conversation, and not only when someone asks.
It must be clear and distinguishable, and meet applicable accessibility requirements. The guidelines are explicit about what does not count: a small snippet of text hidden in your website footer is not enough, and neither is a line buried in your terms and conditions.
So the disclosure belongs in the interface itself. For a chat widget that means a label or an opening line inside the window. For AI telephony it means the AI says so at the start of the call.
There is one exception: if it is already obvious to a reasonably observant user that they are talking to AI, you do not need to say it separately. Leaning on that exception is risky, because you are the one who has to explain afterwards why it was obvious.
The next deadline: 2 December 2026
A second clock is running, and it is often missed.
For generative AI systems already on the market before 2 August 2026, the Digital Omnibus grants an extension until 2 December 2026 to meet the machine-readable marking requirement for generated content.
Content generated and published before that date does not need to be labelled retroactively. This is about what you produce from then on.
For most smaller businesses buying a chatbot or AI telephony from a supplier, this is mainly a question to put to that supplier: are you handling this, and by when?
What else the Digital Omnibus changed
The Digital Omnibus entered into force on 27 July 2026, following publication in the Official Journal of the EU on 24 July 2026.
Its main effect lands on high-risk systems, not on chatbots. The obligations for the Annex III high-risk category were postponed from 2 August 2026 to 2 December 2027.
For customer service that is reassuring news: the heavy obligations moved further away, while the light transparency obligation that does apply to you came into force as planned.
In short: the rule that governs your chatbot was not delayed. The rules that were delayed probably are not about you.
How large are the fines?
Breaching the Article 50 transparency obligations carries a maximum of 15 million euro or 3 percent of worldwide annual turnover, whichever is higher.
That is a different figure from the one usually quoted. The AI Act's heaviest penalty, 35 million euro or 7 percent, is reserved for prohibited AI practices. A customer service chatbot does not fall into that category.
In practice the exposure for a smaller business acting in good faith is limited. But the fix is so straightforward that carrying the risk would be a waste.
What to check today
Open your own chat widget as a visitor. Can you tell within a second that you are talking to AI, without clicking or scrolling? If not, that is your first action point.
Call your own business number if you use AI telephony. Does the AI say so at the start of the call? Not as a rushed disclaimer, but clearly enough that an ordinary caller takes it in.
Check the same for your other channels. If you run a WhatsApp agent, or AI replies to your LinkedIn or Instagram messages, the obligation applies there just as much.
Write down what you did. A short document listing which AI systems you use, how you inform users and since when is enough. It does not need to be a binder, but you want it if questions ever come.
Ask your supplier about marking of generated content ahead of the 2 December 2026 deadline.
How this works in Heyloha
In Heyloha you decide how the AI chatbot and AI telephony open the conversation: you set the greeting yourself, per channel. So check that it makes clear from the first line that people are talking to AI, and say it plainly.
It pays off in the conversation too: people answer more openly and get to the point faster when they know who they are talking to.
To let customers reach a person, switch on handover to your team. On chat channels (website chat, WhatsApp, Instagram and LinkedIn) a colleague takes over with the whole conversation, so the customer does not have to start again. On the phone, you set per colleague when the agent may put callers through: with a warm transfer your colleague first hears why the customer is calling, then accepts the call. If nobody answers and callback requests are on, the agent takes a callback request and sends it to the address you choose.
If you have questions about your specific setup, we are happy to look at it with you. This article is not legal advice, and if you are unsure about your own situation a lawyer is the right person to ask.
A rule you would have wanted to follow anyway
For most businesses the EU AI Act comes down to one question: do your customers know they are talking to AI? If the answer is yes, you are largely there.
And honestly, it is a rule you would have wanted to follow without a law behind it. Customers who think they are talking to a person and find out halfway through that they are not do not come back.
Want AI customer service that answers first and can bring in your team when a customer wants a person? Try Heyloha free for 14 days. Chat is live within about 30 minutes.
Frequently asked questions about the EU AI Act and customer service
Since when do the EU AI Act rules apply to chatbots?
Since 2 August 2026. From that date the Article 50 transparency obligations apply to AI systems that interact directly with people, including chatbots and AI telephony. The rule applies regardless of when the system was placed on the market.
What must my chatbot disclose to comply with the AI Act?
That the user is interacting with AI, at the latest at the first interaction, clearly and distinguishably within the interface itself. A line of text in the footer or a mention in your terms and conditions does not satisfy the European Commission's guidelines.
What happens on 2 December 2026?
The extension expires for generative AI systems already on the market before 2 August 2026 to meet the machine-readable marking requirement for generated content. Content published before that date does not have to be labelled retroactively.
How big is the fine for breaching Article 50?
Up to 15 million euro or 3 percent of worldwide annual turnover, whichever is higher. The frequently quoted maximum of 35 million euro or 7 percent applies only to prohibited AI practices, not to transparency in customer service.
Has the AI Act been delayed?
Partly, but not the part covering chatbots. The Digital Omnibus postponed obligations for Annex III high-risk systems to 2 December 2027. The transparency obligation for chatbots and AI telephony took effect as planned on 2 August 2026.