AI Act Article 52 requires that individuals be informed when they're interacting with an AI system. In customer service contexts, this sounds straightforward but has some operational complexity:
**What we implemented:**
- **Explicit disclosure at session start:** "You're now connected with an AI assistant. A human agent is available if you'd prefer." This appears before the first AI message, not buried in terms.
- **Persistent indicator:** A subtle badge in the chat UI that says "AI Assistant" remains visible throughout the conversation.
- **Escalation path:** Users can type "human" or click a button at any point to transfer to a human agent. The handoff includes the conversation transcript.
**The trickier question is indirect AI interaction:** When a customer calls a hotline and an AI transcribes, routes, and suggests responses to the human agent — is the customer "interacting with" the AI? Our legal counsel took the position that since the customer is speaking to a human who makes all decisions, Art. 52 disclosure isn't triggered. But this is a gray area.
**Documentation burden:** We maintain an internal register of all AI systems used in customer-facing contexts, noting the Art. 52 disclosure mechanism for each. This doubled as useful documentation for our ISO 27001 audit.
**Vendor-supplied AI:** If you're using a third-party chatbot (Intercom, Zendesk AI), check whether *they* provide the Art. 52 disclosure or if it's your responsibility as the deployer. We found most vendors don't cover this — it falls on you.
Has anyone dealt with the "emotion recognition" disclosure requirement under Art. 52(2)? That one seems much more invasive and I'm trying to understand practical compliance approaches.