A company wants to add a mechanism to automatically suppress outputs that contain harmful expressions or undesirable topics before publicly launching a chatbot built with Bedrock. Which feature is MOST appropriate for this requirement?
Identify the Bedrock feature that suppresses harmful content.
Amazon Bedrock Guardrails
This is correct. Amazon Bedrock Guardrails is a safety feature that uses content filters, denied topics, sensitive information filters, and more to automatically suppress outputs containing harmful expressions or undesirable topics.
Amazon SageMaker Clarify
This is incorrect. SageMaker Clarify is a service for detecting bias in data/predictions and visualizing explainability. It does not automatically suppress harmful content from generative AI outputs.
Amazon Macie
This is incorrect. Amazon Macie is a service that detects sensitive data (such as PII) in Amazon S3. It does not automatically suppress harmful content from generative AI outputs.
Amazon Augmented AI (A2I)
This is incorrect. Amazon A2I is a service that integrates human review workflows for AI predictions. While human review can be used to address harmful outputs, the requirement specifies automatic suppression, which A2I cannot fulfill on its own.
When the question involves suppressing harmful content, the answer is almost always Amazon Bedrock Guardrails. It controls model outputs using content filters, denied topics, sensitive information filters, and grounding checks. Clarify (bias detection), Macie (sensitive data detection in S3), and CloudWatch (monitoring) each serve different purposes.