In this article8 sections
The rule that catches most SMEs off guard
AI chatbot EU AI Act compliance is not optional. Under Article 50 of the EU AI Act, providers must design AI systems that interact directly with people so that those people are told they are talking to an AI. As the deployer, you choose the vendor, configure the bot and must not switch that notice off. The requirement applies whether your chatbot handles customer support tickets, answers employee HR queries, or qualifies inbound sales leads.
Article 50 has applied since 2 August 2026. Procurement decisions made today lock in your compliance posture for months, so check it before you sign.
What Article 50 actually requires
Article 50(1) requires providers to design chatbots so people are told they are talking to an AI. Article 50(5) adds how:
-
Timely disclosure. The information must be given at the latest at the first interaction, not buried in a footer or terms page.
-
Clear and distinguishable notice. The disclosure must be clear and distinguishable. A single line of greyed-out 9pt text does not meet this standard.
Offering a route to a human is good practice, not an AI Act requirement. Separate rules cover AI-generated content: deep fakes and AI-written text published to inform the public must be disclosed (Article 50(4)).
There is one limited exception: no notice is needed where it is obvious "from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use" that they are dealing with an AI. Relying on this exception is risky without documented reasoning.
Penalties for non-compliance with Article 50 can reach €15 million or 3% of global annual turnover, whichever is higher, under Article 99. For SMEs the lower figure applies, so for an SME with €10 million in turnover the ceiling is €300,000. National authorities set the actual penalty.
Three chatbot scenarios, three compliance gaps to close
Customer-facing support bot
You embed a chatbot on your website to handle first-line support queries. This is the highest-visibility deployment. The gap most SMEs miss: the disclosure notice appears only after the user types their first message. That is too late. Article 50 requires the information at the latest at the first interaction. A simple, visible banner stating "You are chatting with an AI assistant" displayed when the chat window opens is sufficient, provided it is readable and not dismissible before the user sees it.
Internal HR bot
An HR bot that answers employee questions about leave entitlements, payroll queries, or disciplinary procedures carries additional obligations beyond Article 50. If the system influences decisions related to employment, it may also qualify as a high-risk AI system under Annex III of the AI Act, specifically category 4 (employment, workers management, access to self-employment). If it is high-risk, from 2 December 2027 you need human oversight, logging and worker information (Article 26), and the vendor must supply a declaration of conformity and instructions for use (Articles 13 and 47). Article 50 disclosure is the floor here, not the ceiling.
Sales qualification bot
A bot that scores or routes inbound leads is less likely to trigger high-risk classification, but the Article 50 disclosure obligation still applies in full. The additional risk here sits at the intersection of the AI Act and the AVG (GDPR): if the bot collects personal data during the conversation, you need a lawful basis, a retention policy, and a record of processing activity. The European Commission's guidance on AI and data protection confirms that the AI Act and GDPR operate in parallel, not as alternatives.
The vendor due diligence checklist
Before signing any contract for a chatbot or conversational AI tool, run every vendor through these twelve questions. Document the answers. If a vendor cannot answer clearly, that itself is a compliance signal.
Disclosure and transparency
-
Does the system display an AI disclosure notice before or at the start of every interaction, by default?
-
Can the disclosure text be customised to match your brand while remaining compliant?
-
Does the system support a clear escalation path to a human agent? (Good practice, not a legal requirement.)
-
Does the vendor provide template disclosure language that satisfies Article 50?
Risk classification
-
Has the vendor confirmed in writing whether the system is classified as high-risk under Annex III?
-
If high-risk: will an EU declaration of conformity be available when the high-risk rules apply (2 December 2027)?
-
Does the vendor supply the instructions for use required by Article 13, so you can meet your duties under Article 26?
Data and GDPR
-
Where is conversation data stored, and for how long?
-
Is the vendor acting as a data processor under the AVG, and is a Data Processing Agreement in place?
-
Does the system support data subject access and deletion requests?
Incident and audit readiness
-
Does the vendor maintain logs that you can access in the event of a regulatory inquiry?
-
What is the vendor's process for notifying you of material changes to the AI model?
Print this list. Use it as a literal attachment to your procurement process. Any vendor that refuses to answer these questions in writing should not make it past the shortlist stage.
What a compliant Article 50 notice looks like
The notice does not need to be long. It needs to be:
-
Visible before or at the moment the user begins typing
-
Plain language, not legal boilerplate
-
Accurate, meaning it correctly describes the nature of the system
-
Persistent, available throughout the session, not just at the start (good practice)
A working example for a support bot: "Hi! I'm an AI assistant. I can answer most questions about your order, but you can ask for a human at any time by typing 'agent'."
For an internal HR bot, add the scope boundary: "I'm an AI assistant trained on [Company] HR policies. My answers are informational. For decisions affecting your employment, please speak with an HR team member."
These notices serve a dual purpose. They satisfy Article 50, and they manage user expectations in a way that reduces escalation volume.
AI literacy: the obligation nobody talks about
Article 50 gets the attention, but Article 4 is equally relevant for deployers. It requires companies to take measures to support the AI literacy of staff who work with or oversee AI systems, and it has applied since 2 February 2025. No specific level is required. For an SME, the natural measure is role-appropriate training for anyone who manages the chatbot configuration, reviews its outputs, or handles escalations. Keep your training record current: it is your evidence of the measures you took.
For most SMEs, this obligation translates to two practical steps: a brief AI literacy module for the relevant team, and a simple training log that records who completed it and when.
Your next step
Vendor due diligence is not a one-afternoon job, but it is a manageable one when you have the right framework. Start with the twelve-question checklist above. Then map your existing chatbot deployments against the three scenarios described here to identify which ones need urgent attention and which need deeper high-risk assessment.
If you are unsure where your current tools sit, Khairos Comply offers a free 2-minute compliance check that flags your highest-priority obligations under the AI Act. Run it before your next vendor renewal.
References
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.