Search/MCP results can surface secrets, PII, or content above a caller's clearance. Add data-loss prevention on the response payload: classify indexed content by sensitivity tier, filter/redact results the caller isn't cleared for (the data_classification dimension of the authz decision), redact secret-pattern snippets, and emit an audit event on redaction. Prereq: MCP authz + audit (GCP phase 4); ideally the in-house policy engine.
Follow-up from design doc docs/plans/2026-06-09-enterprise-hardening-and-cloud-deployment.md (enterprise hardening, GCP-first).
Local TODO: .planning/todos/pending/2026-06-09-enterprise-dlp-classification-aware-redaction.md
Search/MCP results can surface secrets, PII, or content above a caller's clearance. Add data-loss prevention on the response payload: classify indexed content by sensitivity tier, filter/redact results the caller isn't cleared for (the data_classification dimension of the authz decision), redact secret-pattern snippets, and emit an audit event on redaction. Prereq: MCP authz + audit (GCP phase 4); ideally the in-house policy engine.
Follow-up from design doc
docs/plans/2026-06-09-enterprise-hardening-and-cloud-deployment.md(enterprise hardening, GCP-first).Local TODO:
.planning/todos/pending/2026-06-09-enterprise-dlp-classification-aware-redaction.md