Version 1.41.0
8 days ago by Tech admin
Guardrails
- New guardrail — PHI detection. Detects protected health information in a prompt — patient names, clinical details (conditions, medications, procedures), and health identifiers — and redacts it before the prompt reaches the model, or blocks the request. Unlike pattern-based PII redaction, it uses a machine-learning model that reads context, so it catches free-text health information. Selectable categories and a confidence threshold.
Security
- Vulnerability fixes.