AIF-C01 Practice Question: Security, Compliance, and Governance for AI Solutions

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Amazon Comprehend for toxicity detection

  Why it's wrong here

  Comprehend performs post-hoc text analytics such as sentiment and toxicity scoring; it cannot intercept or block Bedrock generations. Guardrails for Amazon Bedrock applies configurable content filters and denied topics directly to model input and output. Comprehend suits analysing stored documents or transcripts, not enforcing runtime safety policies.

• ✗

  Amazon SageMaker Clarify

  Why it's wrong here

  SageMaker Clarify detects bias and explains model predictions during development; it does not intercept Bedrock inference to enforce runtime content safety. It is tempting because it addresses bias, but Guardrails for Amazon Bedrock is the service that filters harmful content at generation time.

• ✗

  AWS WAF to filter model responses

  Why it's wrong here

  AWS WAF inspects HTTP requests at the web application layer, filtering traffic by IP, headers and URI patterns; it cannot parse generated text for harmful content. Bedrock Guardrails evaluates prompts and completions against content filters and denied topics. WAF is correct for shielding a public web application from exploits such as SQL injection.

• ✓

  Amazon Bedrock Guardrails

  Why this is correct

  Guardrails applies configurable content filters and denied-topic policies directly to Bedrock model invocations, blocking harmful or biased output before it reaches the application. This satisfies the requirement to enforce content safety policies at inference time, which prompt engineering alone cannot guarantee.

Go deeper

Related to this question