A county must forensically review a bad permitting answer to see the exact prompt and model response. Which monitoring capability best supports that analysis?
Select an answer to reveal the explanation.
Short Explanation
When a permitting answer goes sideways, you need the black box recording—not a shrug. Bedrock Model Invocation Logs capture the prompt and response for forensic review.
Full Explanation
Forensic review of a bad permitting answer requires reconstructing the exact prompt and model response. Enabling Amazon Bedrock Model Invocation Logging captures those payloads so county investigators can see what context, instructions, and completion produced the error. That monitoring capability is the authoritative trail for GenAI incident analysis in municipal permitting—far more reliable than informal recollection.
Relying on verbal recollections from the resident who received the answer is incomplete and unverifiable for technical root cause. Enabling only VPC Flow Logs while ignoring FM request payloads shows network metadata, not prompt/response content. Deleting invocation history daily without retaining a review window destroys the forensic evidence the county needs after incidents surface.
Exam caveat: Bedrock Model Invocation Logging is the exam-aligned answer for prompt/response forensics; flow logs and immediate deletion do not substitute. Operational check: enable invocation logging to a secured S3/CloudWatch destination with encryption and least-privilege access, set retention to match incident-response policy, and practice retrieving a sample bad-answer case end-to-end.