311 reviews security logging for a GenAI assistant. Which practice best balances forensics and privacy?
Select an answer to reveal the explanation.
Short Explanation
You need breadcrumbs without spilling resident secrets. Log prompts and responses with privacy redaction so 311 can investigate and still respect privacy.
Full Explanation
311 GenAI security logging should capture prompts and responses with privacy redaction so forensics and privacy both hold. Investigators need enough content to reconstruct abuse, prompt-injection impact, or faulty answers affecting residents; privacy requires stripping or masking direct identifiers and sensitive fields, plus strict access control and retention limits on what remains. Redacted forensic logging is the balanced architecture pattern—not silence and not unrestricted raw retention.
Logging nothing leaves investigations without an evidence trail and weakens accountability. Storing full prompts and responses forever with no redaction or access control creates a secondary trove of resident content ripe for insider misuse or breach. Publishing raw chat logs on the open-data portal without review is disclosure, not transparency done safely.
Exam caveat: the balanced answer is log-with-redaction (and implied access control), not “log everything forever” or “log nothing.” Open-data publication of raw chats fails privacy.
Operational check: implement prompt/response logging with automated redaction rules, role-based access, retention schedules, and periodic access audits; verify a sample incident reconstruction succeeds using redacted logs without exposing unnecessary personal data.