Every SaaS product with a chatbot, AI assistant, or conversational interface has obligations under the EU AI Act. Article 50 creates specific transparency requirements for AI systems that interact directly with people.
What Article 50 Requires
If your AI system interacts directly with natural persons, you must ensure that the person is informed they are interacting with an AI system. This applies to:
- Customer support chatbots
- AI assistants embedded in products
- Virtual agents handling inquiries
- Any conversational AI that could be mistaken for a human
The disclosure must happen in a timely, clear, and intelligible manner — before or at the start of the interaction.
What About AI-Generated Content?
Article 50 also covers synthetic content generation:
- Text generation: Content produced by AI must be machine-readable as AI-generated
- Audio/video: AI-generated or manipulated audio, images, or video (deepfakes) must be labeled
- Emotion recognition: Systems detecting emotions must inform users they're being analyzed
The "Obviously AI" Exception
Disclosure is not required when the AI nature of the system is obvious to a reasonable person. However, relying on this exception is risky — what seems obvious to your engineering team may not be obvious to your users.
Practical Implementation
For Chatbots Add a clear indicator at the start of the conversation: - "You're chatting with an AI assistant" - A visible AI badge or label in the chat interface - An initial message disclosing the AI nature
For Generated Content - Add metadata tags indicating AI generation - Include visible labels on AI-generated images or text - Maintain logs of what was generated and when
Common Mistakes
- Burying the disclosure — Putting "powered by AI" in the footer doesn't count. It must be prominent and timely.
- Disclosing only once — If users can forget they're talking to AI (e.g., long conversations), periodic reminders may be appropriate.
- Assuming users know — Don't assume technical sophistication. Many users genuinely can't distinguish AI from human responses.
Risk Level
Chatbot transparency is a Limited Risk obligation — not high-risk, not prohibited. The requirements are lighter than high-risk AI systems, but non-compliance still carries fines up to €7.5M or 1% of global turnover.
The good news: compliance is straightforward. Add clear labels, maintain documentation, and you're largely covered.