A customer-service chatbot must refuse to discuss competitor products and block hate speech, without retraining the model. Which Bedrock feature enforces these policies at inference time?
A.Bedrock Guardrails — denied topics + content filters
Independent policy enforcement layer. Correct.
B.System prompt instructions only
Prompt instructions are advisory — prompt injection can bypass them.
C.Fine-tuning the model on refusals
Retraining for each policy is heavyweight — guardrails are runtime rules.
D.IAM policies on the API
IAM controls WHO calls the API — not WHAT it generates.
Why this answer
Guardrails enforce policies independent of the model: denied topics (e.g., competitors), content filters (hate/violence/etc.), PII redaction, word filters — applied to prompts and responses. Prompts can be bypassed; guardrails can't.
Correct answer: A · Last reviewed 2026-11-05
Official AWS references
Practice this interactively
Answer without seeing the solution first in the Responsible AI practice session.
Practice Responsible AI