A municipal lamp-shop helper must not return abuse, or leak a crew member’s personal cell number, even if a caller asks. Which security control belongs on those outputs?
Select an answer to reveal the explanation.
Short Explanation
Think of a helper that must not return abuse or leak a crew cell number, even if asked. Amazon Bedrock Guardrails is the output filter. A principles essay does not sit on the live answers.
Full Explanation
The municipal lamp-shop helper must not return abuse or leak a crew member's personal cell number even when a caller asks. Amazon Bedrock Guardrails, or an equivalent output filter, is the security control that sits on those outputs. Filters and denied-topic or PII policies catch unsafe or sensitive completions before they reach the public desk and become a personnel-privacy incident.
A Domain 4 principles essay with no filter on live outputs fails because written principles do not intercept runtime responses. Amazon QuickSight-style charts fail because dashboards do not redact cell numbers in generative answers. From-scratch pre-training fails because a weekend crawl is not an output filter on production completions and does not stop abuse text at inference time.
Exam caveat: input and output guardrails complement—but do not replace—data minimization in the knowledge base; if personal cell numbers are indexed, leakage risk remains higher even with filters in place. Operational check: enable Guardrails (or equivalent) for abusive content and PII on the lamp-shop helper, test with requests that solicit crew cell numbers, verify blocks or redactions, and remove personal phones from grounding documents where they do not belong.