AIF-C01 Practice Question: Using Amazon Bedrock to deploy a generative AI…

Answer analysis

Option-by-option breakdown

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

• ✗

  Enable data privacy settings in the model invocation

  Why it's wrong here

  Data privacy settings govern how inputs and outputs are handled and retained, not whether content is harmful; they cannot block offensive output. They are tempting because they are configured at invocation, but content filtering requires a guardrail with denied topics and content filters attached to the model.

• ✗

  Use prompt engineering to instruct the model to be safe

  Why it's wrong here

  Prompt engineering only influences model behaviour probabilistically; it cannot enforce a deterministic block on harmful output. It is tempting because instructions are quick to add, but Bedrock Guardrails applies configurable content filters and denied topics that intercept offending responses regardless of prompt wording.

• ✗

  Set up invocation logging to review all outputs

  Why it's wrong here

  Invocation logging records prompts and responses for audit and troubleshooting after generation; it does not intercept or block harmful content. It is tempting because logs support review, but prevention needs a guardrail policy that filters content before the response reaches the user.

• ✓

  Configure denied topics and content filters

  Why this is correct

  Denied topics define subject areas the model must refuse, while content filters block harmful or offensive categories such as hate, violence and sexual content, together satisfying the requirement to prevent harmful output. Both are configured as guardrail policies applied to the model invocation.

Go deeper

Related to this question