What

Add an operator to create Amazon Bedrock guardrails for implementing safety controls in generative AI applications.

Use Case

Teams deploying LLM-powered applications need to enforce content safety policies (block hate speech, filter PII, deny off-topic discussions) as part of their deployment pipeline. A guardrail is created with content filters and topic policies, then associated with the Bedrock model. Having this in a DAG makes safety policy changes versioned, auditable, and promotable through dev/staging/prod:

create_guardrail >> configure_model_endpoint >> run_eval_tests >> notify_team

The create_guardrail API creates a DRAFT version that is immediately usable — no separate version step needed for dev/test workflows.

How

• New BedrockCreateGuardrailOperator using AwsBaseOperator[BedrockHook] pattern
• Supports content filters, denied topics, word filters, PII masking, and contextual grounding
• Idempotent: if_exists="skip" handles ConflictException
• Uses prune_dict for clean kwargs building
• Returns the guardrail ID

Files

• operators/bedrock_guardrail.py — operator
• test_bedrock_guardrail.py — unit tests (4 tests)
• example_bedrock_guardrail.py — system test with teardown
• bedrock_guardrail.rst — docs
• provider.yaml + get_provider_info.py — updated

Testing

• Unit tests: 4 passed
• mypy: no issues
• Static checks: 43 passed, 0 PR-related failures