Talking to a bot? You have the right to know
Author: Andrzej Albera, Responsible for strategy and the company's direction
The situation. A customer writes to an online shop's chat in the evening. “Anna” replies: quickly, politely, without typos. After a quarter of an hour the customer asks to speak to a manager and learns that “Anna” is a program. She feels deceived, even though the bot helped her well. The next day the shop gets a bad review, and it may also get a complaint to the regulator.
What the law says
- A bot must introduce itself. Under Article 50 of the EU AI Act, a person interacting with an AI system must be informed about it, unless this is obvious from the context. The information must be clear and given at the latest at the first interaction.
- Deepfakes must be labelled. Images, video or audio generated or manipulated by AI that resemble real people, places or events must be disclosed as AI content. We describe how this works in newsrooms and agencies on our page Media and communications.
- So must texts on matters of public interest. AI-generated text published to inform the public on matters of public interest must be labelled, unless it has undergone human editorial review and someone takes responsibility for it.
- The deadlines have not moved. The Digital Omnibus package postponed obligations for high-risk systems, but not Article 50. Transparency obligations have applied since 2 August 2026.
- Fines: up to EUR 15 million or 3% of global turnover. Small and medium-sized companies pay the lower of the two.
- Complaints: since 28 October every customer can file a complaint with Poland's AI regulator, the Commission for the Development and Safety of Artificial Intelligence (KRiBSI). Read more about the act and its deadlines on our page The Polish AI Systems Act.
What to do
- List the places where customers meet AI. The website chat, a hotline with a synthetic voice, automatic email replies, images in adverts. There are often more of them than the board thinks.
- Let the bot introduce itself in the first sentence. For example: “Good evening, I am the shop's AI assistant. If you would rather talk to a consultant, just type: consultant”. A route to a human is not a legal requirement, but customers appreciate it.
- Label AI content. Images, recordings and voices that imitate real people get a visible “generated by AI” label. Texts published on behalf of the company go through a person who is responsible for them.
Our SAIE ecosystem helps with this. PROXY:AI keeps an AI usage log, so it shows every place where the company uses models, including those where AI talks to customers. TWIN:DESK is a Digital AI Desk where assistants answer from the company's knowledge and give the source of every answer. The CDF methodology starts with an initial assessment in which we list every point where customers meet AI.
Frequently asked questions
Must every chatbot say it is a bot?
Yes, unless it is obvious from the context to a reasonably attentive person. When in doubt, it is better to say so. One sentence costs less than a complaint.
Must every email written with AI be labelled?
No. The duty to label text applies to publications on matters of public interest and does not apply to texts edited by a person who is responsible for them. Ordinary correspondence with customers does not need a label, but data protection rules still apply to it.
Didn't the Digital Omnibus postpone these deadlines?
No. It postponed obligations for high-risk systems until December 2027. The duty to disclose AI interaction and to label deepfakes has applied since 2 August 2026. Providers of content generation tools placed on the market before that date have until 2 December 2026 for technical marking.
What is the penalty for a bot that does not introduce itself?
A fine of up to EUR 15 million or 3% of turnover, the lower of the two for small and medium-sized companies. Usually a customer complaint and an inspection come first. We wrote about how to prepare for one in AI inspections and fines from 28 October.
Can a bot have a name, such as “Anna”?
It can, as long as it is clear from the start that it is an AI assistant. A name is not the problem. Pretending to be human is.
The opportunity
A quick answer from a bot at 10 pm can be better for a customer than waiting until morning for a consultant. On one condition: the customer knows who they are talking to. A bot that says plainly what it is and knows the company's knowledge becomes the company's calling card. A bot pretending to be human leads straight to a bad review.