Sample data. Real workflow.
A prompt with sensitive data meets the policy engine
Pick a sample prompt. Watch what the engine detects, which policy fires, what the model is allowed to see, and what gets logged. Then switch enforcement off.
No customer data is shown. Every name, record and value below is synthetic. This follows the documented AiT AI Gateway and Hosted Agents flow: policy check first, approved model second, audit log always.
1. Prompt as typed
Summarize this intake note for the care team: Jane Q. Sample, DOB 01/02/1970, MRN TEST-000001, reports poor sleep for three weeks.
2. Detection
- Patient name
Jane Q. Sample - Date of birth
01/02/1970 - Medical record number
TEST-000001
3. Policy that fired
Protected health information is removed before any prompt leaves the tenant boundary. Action: redact.
4. Redacted prompt
Summarize this intake note for the care team: [PATIENT_NAME], DOB [DOB], MRN [MRN], reports poor sleep for three weeks.
5. Model routing
Approved model via the AI gateway, de-identified payload only
6. Audit-log entry
- event
- sample-0001
- user
- user@example.test
- policy
- POL-PHI-01
- detections
- 3
- action
- redact
- stored
- redacted copy only
- entry
- sha256:3b1f...c0de (sample)