How to Build a PHI-Redacted Audit Trail for FM Prompts

Why the Answer Is Correct

Option C is the only choice that satisfies every clause of the requirement set at once. S3 Object Lock in compliance mode creates write-once-read-many storage, so the prompt-completion audit trail cannot be altered or purged early, which is exactly what an audit trail of all prompts and completions demands. Tag-based S3 Lifecycle policies give the company an enforcement point for healthcare-specific retention periods applied to PHI-tagged objects, and the Lambda function redacts patient information after processing so that stored records carry no PHI. Because redaction happens as part of processing rather than at the end of a retention window, the company still tracks each prompt-completion pair while keeping persistent records clean.

Why the Other Options Are Wrong

Option A leaves PHI intact in S3 until a retention period expires and then deletes it, which violates the explicit requirement to redact any PHI from stored records, and default server-side encryption is a transport-at-rest control rather than a redaction or retention mechanism. Option B leans on Amazon Bedrock Prompt Management and Flows to detect PHI, but those services orchestrate prompts and pipelines rather than perform clinical PHI detection, and Guardrails sensitive-information filters operate on model interactions instead of guaranteeing redacted stored logs plus enforced healthcare retention. Option D puts patient data into an Amazon Bedrock knowledge base, which is a retrieval corpus for RAG rather than an immutable prompt-completion audit log, and a retention Lambda without Object Lock cannot prevent records from being modified. None of A, B, or D delivers compliance-mode immutability alongside redaction and lifecycle-based retention.

Community Comment Notes

A single commenter, TNT87, argued for option B, repeating the option wording that the team should "Use Amazon Bedrock Prompt Management and Amazon Bedrock Flows to detect PHI" and closing with "Ans B." That reasoning addresses masking during inference but never establishes an immutable audit trail or a healthcare retention mechanism, which are explicit requirements in the stem. The comment also assumes Guardrails can enforce retention on S3 logs, when the option only offers generic S3 retention settings. On the exam, weight the words compliance mode, redact, and retention policies, since those three together point to option C.