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.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
"""Wave 4c — self-improving detections: feedback -> tuning suggestions."""
from __future__ import annotations
from app.services.detection_tuner import DispositionRecord, suggest_tuning
def _fp(rule, entity=None, human=False):
return DispositionRecord(rule_id=rule, disposition="false_positive", entity=entity, human_verified=human)
def _tp(rule):
return DispositionRecord(rule_id=rule, disposition="true_positive")
def _btp(rule):
return DispositionRecord(rule_id=rule, disposition="benign_true_positive")
def test_low_fp_rule_gets_no_suggestion():
records = [_tp("r1")] * 8 + [_fp("r1")] * 2 # 20% FP
assert suggest_tuning(records, min_samples=10) == []
def test_entity_clustered_fps_suggest_exception():
records = [_fp("r1", entity="scanner-01") for _ in range(9)] + [_fp("r1", entity="other")] + [_tp("r1")] * 2
out = suggest_tuning(records, min_samples=10)
assert len(out) == 1
assert out[0].action == "add_exception"
assert out[0].entity == "scanner-01"
def test_diffuse_fps_suggest_lower_severity():
records = [_fp("r2", entity=f"h{i}") for i in range(12)] + [_tp("r2")] * 3
out = suggest_tuning(records, min_samples=10)
assert out and out[0].action == "lower_severity"
def test_overwhelming_human_confirmed_fps_suggest_suppress():
records = [_fp("r3", entity=f"h{i % 8}", human=True) for i in range(40)]
out = suggest_tuning(records, min_samples=10)
assert out and out[0].action == "suppress"
assert out[0].requires_approval is True
def test_benign_true_positive_not_counted_as_fp():
# 10 BTP (valid detections) + 0 FP -> the rule is NOT tuned into silence.
records = [_btp("r4") for _ in range(10)] + [_tp("r4")] * 2
assert suggest_tuning(records, min_samples=10) == []