Skip to main content
The EU AI Act places two direct obligations on real-time conversational AI:
  • Biometric emotion recognition is prohibited in workplace and education contexts — a system may not infer emotion from what it sees and hears (facial expression, tone of voice). Emotion the LLM infers from what a user says is not covered by the ban.
  • AI disclosure is required — participants must be clearly told they are interacting with an AI, at the latest at the first interaction.
Tavus surfaces two building blocks so you can meet both obligations without changing PAL prompts or client code:
  1. A conversation-level policy parameter you set to eu for European participants.
  2. PAL-level fields (emotion_recognition, disclosure_type, verbal_disclosure, visual_disclosure) that either follow policy automatically or can be pinned to a specific behavior.

​
The policy parameter

Set on POST /v2/conversations:
When policy is eu, every PAL layer set to auto (the default for the fields below) switches to its EU-safe behavior for the duration of that conversation. Layers with an explicit setting are honored as-is — policy never overrides an explicit choice. You choose the value per conversation. Tavus does not geolocate the caller; passing the correct value is the customer’s responsibility.

​
Biometric emotion recognition

Configured on the perception layer via emotion_recognition:
emotion_recognition only touches biometric signals. The LLM can still reason about emotion from what the user says, which the ban does not restrict.

​
Google Meet / Zoom PALs

Google Meet / Zoom PALs join calendar-invited meetings, where the meeting organizer is not necessarily an EU participant. If your Google Meet / Zoom PAL is used strictly outside EU workplace and education contexts, you can pin it to full:

​
Informed AI disclosure

Configured at the top level of the PAL:
  • disclosure_type — always, auto (default), or off.
  • verbal_disclosure — sentence the PAL speaks before its greeting. When empty, falls back to Tavus’s default: “Just a note, I am an AI system, not a person.”
  • visual_disclosure — on-screen banner shown while the verbal disclosure plays. When empty, falls back to Tavus’s default: “You are interacting with an AI system.”
Example PAL that discloses automatically for EU conversations:

​
End-to-end example

Create the PAL once with auto defaults, then flip compliance on per conversation by passing policy: "eu":
Create PAL
Create EU-safe conversation
  • Perception layer — full reference for emotion_recognition and Raven-1 audio and visual analysis.