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.
Walkthrough — phishing-payload
Click-through to a credential-harvest page. A phishing email passes perimeter filters because the sender domain was less than 14 days old and had valid DMARC alignment for a typosquatted root. Bob clicks the link 6 minutes after delivery and arrives at a domain that imitates the corporate SSO. The credential POST is captured at the proxy.
- Fixture:
examples/alerts/phishing-payload.json - MITRE techniques: T1566, T1566.002
- Severity at intake:
high - Confidence band (sandbox): 87/100,
high
Run it
# Offline simulator:
aisoc-sandbox demo --scenario phishing-payload
# Real stack (after `make up`, with `pip install -e packages/aisoc-cli`):
aisoc submit examples/alerts/phishing-payload.json
What the agent does, step by step
Step 0 · DetectAgent · detect
The detection chain fires three rules in sequence: domain-young
(domain age < 14 days), email-spf-pass-but-typosquat (DMARC passed
because the typosquat owns its own SPF), and proxy-credential-post
(form fields username + password posted to a non-allowlisted
domain). Fusion correlates the three events into one incident with a
deterministic correlation narrative.
Step 1 · TriageAgent · triage
The triage rationale references the known phishing playbook: the click happened inside the standard credential-harvest window from email delivery (under 10 minutes), and the destination domain registered less than 14 days ago — both are public indicators of campaign-style credential phishing.
Decision: Confidence high (87/100).
Step 2 · HuntAgent · hunt
Pivots on Bob's identity to see whether anyone else in the org
received email from acme-paychecks.com in the last 7 days. The hunt
also checks whether Bob's session shows any post-credential-POST API
calls (the attacker hasn't logged in with the stolen credential
yet — there's still a window to revoke).
Step 3 · RespondAgent · respond
Proposes the canonical phishing containment:
proxy.block_domain({"domain": "acme-paychecks.com"})user.session.revoke_all({"user": "[email protected]"})user.require_password_reset({"user": "[email protected]"})email.search_and_purge({"sender": "[email protected]"})case.create(...)at severityhigh
Decision: 5 actions proposed; awaiting analyst approval (severity high).
What the analyst would do next
- Approve
proxy.block_domainanduser.session.revoke_allimmediately — both have small blast radius. - Run the
email.search_and_purgeaction with caution: it will remove the message from every mailbox in the org. Confirm with the IT lead first. - Add
acme-paychecks.comto the perimeter blocklist permanently, not just at the proxy. - Open a brief learning case from the console's Threat Intelligence
tab (
/threat-intel) so the org's analysts know to look for similar typosquat patterns over the next 30 days.