PII Redaction
PII redaction patterns to prevent data leakage in LLM workflows.
Published:
Admin User
Updated:
published
PII Redaction
PII redaction prevents accidental exposure of sensitive data in prompts, retrieval, and logs.
Enterprise approach: enforce redaction at defined boundary points and keep evidence.
See also
Data Classification Policy & Data Boundaries (Playbook) Privacy Incident RunbookFAQ
Where should redaction happen?
At ingress (before prompts), at retrieval, and before logging/output where needed.
How do we prove redaction works?
Automated tests + audit logs + periodic sampling with evidence packs.
What’s a common failure mode?
PII leaks into logs or prompts due to missing boundary enforcement.
Does redaction reduce quality?
Sometimes; mitigate with task redesign and safer data sources.
What’s the first improvement?
Implement redaction at a single enforced boundary and add tests.