Cyclical Enumerate-Exploit-Enumerate Loop on Single Target — Autonomous Agent Behavior
Detects the distinctive operational loop of autonomous AI attack agents: repeated cycles of enumeration activity followed by exploitation attempts followed by further enumeration — all targeting the same host or subnet. This cyclical pattern reflects the planner-worker architecture where exploitation results feed back into the exploration graph, triggering new rounds of discovery. Human operators may enumerate, exploit, and move on. Autonomous frameworks like ARTEX loop: scan, attempt exploit, re-scan to discover what changed (new access, new services exposed, new credentials available), attempt more exploits with updated context, re-scan again. Each cycle produces a burst of network discovery events (port scans, service probes) followed by authenticated access attempts, then another burst of discovery from the newly accessed host — repeating 3+ times within a session. This detection requires correlation of network connection events with authentication events on the same host or subnet, tracking the oscillation between discovery and access phases. BQL (Bloo Query Language): stream=ep-network where action='NETWORK_CONNECTION' and DstPort in (445, 135, 139, 22, 3389, 5985, 5986, 389, 636, 88) and DstIP rlike '^(10\.|192\.168\.|172\.(1[6-9]|2[0-9]|3[01])\.)' | groupby system, DstIP | having count > 15
Sigma
