Redaction detects and masks sensitive information before it enters an LLM context or is exported in an observability trace. It reduces exposure of personal data in prompts, memory, and telemetry. section 8.3.4 section 9.1.1
Redaction operates at two points in an LLM system:
- During context assembly, mask sensitive entities before they enter the context window or are sent to an external LLM. section 9.1.1
- Before exporting traces, apply field-level masking to prompt payloads; limit trace retention as an additional privacy control. section 8.3.4 section 9.1.1
Customer support context
Before constructing a prompt, filter account identifiers, email addresses, payment details, past tickets, and free-text messages according to purpose and authorization. Tools such as Microsoft Presidio or OpenDP can detect, redact, anonymize, or transform PII before dispatch to an external LLM or persistence in long-term memory. section 9.1.1

