Live attacksiWhat: attacks the supervisor caught against your agent in real time. Different from policy denies — these are detected behaviors (prompt injections, jailbreaks, PII exfiltration attempts, runaway loops) regardless of which action_type fired them.
When: something here means the supervisor blocked or escalated an attempted attack before your agent ran the action. Quiet feed = nobody is poking your agent yet.
Action: click an attack to see the full payload + which detector caught it. If a detector flags too many false positives, tune sensitivity or add an exception.
Levels: critical = likely attack · warn = soft signal · info = informational only.
Footnote: detectors are mapped to the OWASP LLM Top 10 (LLM01 prompt injection, LLM02 PII, LLM06 jailbreak, LLM10 unbounded consumption) — visible in each row for compliance reporting.
Attacks the supervisor caught against your agent in real time — prompt injection, jailbreak, PII exfiltration, runaway loops, and more.
0 Critical
0 Warn
0 Info
401 Unauthorized: {"detail":"integration not found or revoked"}