Transit ops assistants use generative AI. Which logging practice best supports security monitoring without overexposure?

Implementing prompt and response logging with access controls for transit ops assistants supports security monitoring without overexposure: investigators need reconstructable trails of what was asked and answered, while unrestricted access would expose sensitive operational content. The correct practice works because controlled logging enables detection of abuse, leakage, and unsafe suggestions, and access controls enforce least privilege on that high-value log store. Retaining sampled prompt summaries with access controls for ops assistants fails because it provides only sampled visibility rather than a complete prompt/response trail, so security monitoring misses interactions even though access controls limit exposure. Storing encrypted response logging with access controls for ops assistants fails because it captures only responses without the prompts that produced them, so investigators cannot reconstruct what was asked even though access is controlled. Enabling admin activity auditing with access controls for ops assistants fails because it audits system configuration events, not the prompts and responses of transit ops assistants, so it cannot reconstruct AI interactions for security monitoring. Exam caveat: logging should still apply retention limits and may need redaction of secrets or personal data; sampled summaries, response-only logging, and admin auditing are not substitutes for controlled prompt/response logging. Operational check: enable prompt/response logging to a restricted store with role-based access, MFA, and audit trails on log reads; set retention aligned to security and records policy; do not rely on admin activity auditing with access controls for ops assistants as a substitute for prompt/response logging; and verify that security use cases can query the logs during an exercise.