Transit sees anomalous GenAI answers in rider chat. What containment action fits while preserving service?

Anomalous GenAI answers in rider chat indicate a content or model integrity problem in one channel. Containment that fits while preserving service is to disable the GenAI channel while keeping human chat available—stop the harmful automation without stranding riders who still need trip help. That is classic isolate-and-continue for municipal customer service.

Leaving the bad channel up to avoid admitting a problem fails containment and expands harm and reputational damage. Shutting every human and digital channel citywide immediately fails proportionality and continuity; total blackout is rarely required for a GenAI-only anomaly. Ignoring anomalies until riders trend on social media fails detection-to-action discipline and outsources IR triggering to virality.

Exam caveat: containment should be scoped to the impaired AI component while human fallbacks remain. Operational check: one-click or runbook disable for the GenAI rider channel, confirmed human queue capacity, public status language that avoids unsafe silence, and criteria for re-enabling only after root-cause and safety checks. Preserve chat anomaly samples for analysis, rotate any compromised tool credentials if the anomaly suggests prompt injection or connector abuse, and brief transit supervisors so station staff are ready for higher human-chat volume during the disablement window.