Keep PHI out of AI prompts
- Problem
- Staff use assistants to summarize notes and sometimes include patient identifiers.
- What the platform does
- The AI Firewall redacts personal and health information and blocks prompt injection. Discovery finds unsanctioned AI tools.
- Human approval
- Sending sensitive data to an external model requires written approval for that use and data class.
- Evidence
- Usage records showing the caller, the provider tier, and the approval reference.