Parks enables detection for a booking GenAI assistant. Which monitoring focus best fits abuse?
Select an answer to reveal the explanation.
Short Explanation
Weird traffic spikes are a tip-off. Watch inference volume continuously so parks can spot abuse before the booking assistant gets steamrolled.
Full Explanation
Parks’ booking GenAI assistant needs continuous monitoring for anomalous inference spikes that may indicate abuse—credential stuffing of the assistant, automated scraping, prompt-flood denial of wallet/budget, or misuse that degrades resident booking availability. Inference-volume and pattern telemetry, baselined against normal park-reservation behavior, gives operations a chance to throttle, revoke, or investigate before the assistant is steamrolled. Continuous detection completes defense-in-depth after preventive API and auth controls.
Having no telemetry because anomalies are “impossible” is an assumption failure; abuse is routine for public digital services. Monitoring only printer toner levels while ignoring inference traffic measures the wrong system. Turning monitoring off after the first successful demo week abandons production risk management once real residents arrive.
Exam caveat: abuse-oriented monitoring focuses on anomalous inference and misuse signals, not unrelated facility metrics. Detection must remain on after demos.
Operational check: instrument per-client and aggregate inference metrics with spike alerts, tie alerts to playbooks (rate-limit tighten, key revoke, incident bridge), and test that a synthetic flood against the parks assistant triggers detection within the agreed SLA.