EU AI Act Compliance
auraScribe is designed to comply with the EU AI Act's transparency and capability requirements. This page explains how.
AI system identification (Art. 50)
auraScribe uses AI to process meeting recordings. Specifically:
- Transcription: Converting speech to text with speaker diarization
- Speaker profiling: Identifying speakers by name, role, and company from conversation context
- Behavioral analysis: Logging observable conversational signals (tone, pace, hesitations, interruptions)
- Summary generation: Producing structured analysis from transcript and behavioral log
The AI engine is Google Gemini, accessed via Google's Gemini API. Users are informed that AI processes their recordings through the application interface.
Emotion recognition: what auraScribe does NOT do
The EU AI Act restricts emotion recognition systems. auraScribe is not an emotion recognition system. It analyses observable behaviors — what can be heard in the audio — not internal emotional states.
- "Speaker's pace increased during pricing discussion" — observable behavior (compliant)
- "Speaker was anxious about pricing" — emotional inference (non-compliant, automatically rewritten)
A dedicated compliance engine (checkComplianceEU) reviews all AI-generated analysis and automatically rewrites any statements that cross from behavioral observation into emotional attribution.
No biometric categorisation
auraScribe does not categorise individuals based on biometric data for prohibited purposes. Speaker diarization (telling voices apart) is a functional requirement for accurate transcription, not biometric classification.
No profiling for automated decisions
auraScribe does not make automated decisions about individuals. Its outputs are informational — summaries, observations, and coaching suggestions presented to the user for their own interpretation. No scoring, ranking, or automated consequential decisions are made.
Transparency
This page itself is part of our transparency commitment. Users can also review:
- Privacy Policy — full legal details on data handling
- Privacy & Security — technical architecture overview
- Raw Audio Cues Methodology — how behavioral analysis works