PII Scan (Bedrock Guardrails)¶
Uses AWS Bedrock Guardrails to detect and redact personally identifiable information before the model processes input and before responses are returned to callers.
When to use this guardrail¶
Attach to the input position to ensure PII is stripped before it reaches the model.
Attach to the output position to catch any PII the model may have reproduced from
tool results or retrieved knowledge base chunks.
Use both positions for agents that handle customer data in regulated contexts (GDPR, HIPAA).
Recommended agent configuration¶
guardrails:
input:
- ref: "bedrock-pii-scan"
on_fail: "apply" # Apply the redacted version before the model sees the data
output:
- ref: "bedrock-pii-scan"
on_fail: "reject" # If output redaction fails, halt rather than leak PII
Updating the guardrail version¶
When a new version of the Bedrock guardrail policy is published in the AWS console:
- Note the new numeric version (visible in the Bedrock console under Guardrails > Versions).
- Update
guardrail_versionin this file. - Increment
last_updated. - Open a pull request — all agents referencing
bedrock-pii-scanwill automatically pick up the new version after the next compile.
Do not update guardrail_id. This value is immutable once the guardrail is created in AWS.
Relationship to policies.pii_redaction¶
This guardrail and policies.pii_redaction: true can coexist and both run. The policy provides platform-built baseline coverage that never fails; this guardrail adds Bedrock-backed precision and configurability. Use both in depth-in-defense configurations.