Protecting generative-AI inputs and outputs

Review the layered controls used to protect generative-AI inputs and outputs.

Amazon Bedrock Guardrails

Managed controls that help filter or block selected harmful, sensitive, or disallowed model content.

Input control

A check applied before inference to detect unsafe instructions, sensitive data, or invalid request shape.

Output control

A check applied after generation to detect policy violations, harmful content, or sensitive disclosure.

Defense in depth

Using independent controls such as IAM, validation, logging, and review so one failure is not decisive.