Skip to content

armgit / AiSOC

Public Imported from GitHub
0

Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.

3.7 KB
Raw Blame

Walkthrough — lateral-movement

Impossible-travel Okta sign-in. Alice signs in successfully from New York at 09:30 UTC, then again from Saint Petersburg eight minutes later. The geographical pivot is physically impossible inside that window.

Run it

# Offline simulator (< 5 s, no Docker):
aisoc-sandbox demo --scenario lateral-movement

# Real stack (after `make up`, with `pip install -e packages/aisoc-cli`):
aisoc submit examples/alerts/lateral-movement.json

What the agent does, step by step

Step 0 · DetectAgent · detect

Matches the two Okta user.session.start events against the executable detection corpus (2,529 rules as of this writing — docs/detections/truth-table.md carries the generated count). The okta-impossible-travel rule fires on the geographical delta + the sub-10-minute time window, and Fusion lifts the per-alert confidence by another notch because both sessions authenticated through the same FACTOR_PROVIDER.

would-call rules.match({"technique_set": ["T1078", "T1078.004"]})

would-call fusion.score({"window_minutes": 15})

Decision: open alert at severity=high.

Step 1 · TriageAgent · triage

Cross-references the user against Qdrant for prior cases involving [email protected] and walks the Neo4j graph two hops out from the user node to surface devices and groups the user owns. The decision copy ("authenticated session signals look legitimate at the protocol layer, but the geo pivot between sequential events is physically impossible — classic credential takeover") mirrors what an analyst would write at handoff.

would-call qdrant.semantic_search({"k": 5, "collection": "cases"})

would-call graph.neighbours({"depth": 2})

Decision: Confidence high (83/100).

Step 2 · HuntAgent · hunt

Translates the hypothesis "any activity by [email protected] in last 24 h" into an ES|QL query and sweeps the warm tier. The pivot reveals nothing else compromised beyond the alert payload — a clean, contained incident.

would-call nl_to_query.translate({"hypothesis": "any activity by [email protected] in last 24h"})

would-call lake.query({"language": "ES|QL", "row_cap": 1000})

Decision: Pivot complete; no additional compromised entities.

Step 3 · RespondAgent · respond

Proposes the graduated containment: revoke every active session, require password reset, create the case, page the SOC on-call. The blast radius is human_approval_required because the severity is high; nothing executes without analyst confirmation.

would-call user.session.revoke_all({"user": "[email protected]"})

would-call user.require_password_reset({"user": "[email protected]"})

would-call case.create({"title": "Lateral movement — impossible-travel Okta sign-in", "severity": "high"})

would-call case.notify({"channel": "slack", "audience": "soc-on-call"})

Decision: 4 actions proposed; awaiting analyst approval before execution.

What the analyst would do next

  1. Approve user.session.revoke_all (the only single-user-blast-radius action) immediately via the responder PWA.
  2. Pivot on the second IP (203.0.113.50) in the console's Threat Intelligence tab (/threat-intel) to see if it has appeared in other tenants' OTX feeds.
  3. Page Alice's manager from the case to confirm whether she travelled.
  4. If credential reuse is suspected: rotate every API key in Alice's 1Password vault from the credential vault rotation runbook.