Safety & Reliability
Clinical AI must be safe by default. Here's how Relaya protects patients, practitioners, and data.
Clinical Safeguards
Drug Interaction Detection
Real-time flagging when transcribed prescriptions contain potential drug interactions, contraindications, or dosage concerns based on the patient context.
Hallucination Prevention
Linked evidence architecture ensures every AI-generated statement maps to a source utterance. Unsupported claims are flagged, not silently inserted.
Critical Value Alerts
When the AI detects critical lab values, emergency symptoms, or red-flag findings in the consultation, it highlights them prominently in the generated note.
Fail-Safe Architecture
If confidence drops below clinical thresholds, the system flags sections for manual review rather than generating potentially unsafe content.
Our Safety Philosophy
Relaya is an assistive tool, not a replacement for clinical judgement. Our AI is designed to surface information, structure documentation, and flag concerns. but the practitioner always has final authority. We would rather under-generate and flag for review than risk inserting inaccurate clinical content.
Patient Data Protection
- Audio recordings encrypted end-to-end and deleted within 24 hours of processing
- Patient identifiers never used for model training without explicit de-identification
- All AI model outputs logged and auditable for clinical governance requirements
- Automated PHI detection prevents accidental data leakage in exports or integrations
- Regular adversarial testing against prompt injection and data extraction attacks
- Clinical data isolated from all non-clinical systems at the infrastructure level
Questions about safety?
Our clinical safety team is available to discuss our approach with your organisation.
Contact clinical safety team