All alarms
MediumAnomalyMITIGATED
Prompt injection against support model
api-gw-east / user field → AI model · support-llm · updated 2h ago
Summary
A crafted payload attempted to override the system prompt of the customer-support model via a user-controlled field.
Incident timeline
- Detected12:43:29ModelGuard
ModelGuard matched a prompt-override signature in the inbound request body.
- Mitigated12:43:35ModelGuard
Request quarantined at the input firewall; payload never reached the model.
Triage
Set the incident status:
Detection
- Detected by
- ModelGuard
- Category
- Anomaly
- Opened
- 2h ago
Behavior outside the learned baseline.