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
"""#526 — benign_true_positive is a first-class analyst disposition, kept
distinct from false_positive throughout the feedback / rule-tuning surface."""
from __future__ import annotations
from app.api.v1.endpoints.feedback import _VALID_VERDICTS, OverrideSummaryResponse
from app.services.memory_poisoning import PoisoningDetector
def test_feedback_api_accepts_benign_true_positive():
assert "benign_true_positive" in _VALID_VERDICTS
def test_feedback_api_still_accepts_legacy_benign():
# Existing ``benign`` records/verdicts stay valid — no migration required.
assert "benign" in _VALID_VERDICTS
def test_override_summary_tracks_btp_separately_from_fp():
fields = OverrideSummaryResponse.model_fields
assert "benign_true_positive_corrections" in fields
assert "false_positive_corrections" in fields
def test_btp_is_not_treated_as_false_positive_for_rule_suppression():
# The poisoning / re-disposition window keys on FP-like dispositions. A
# benign true positive is a VALID detection: if it landed in this set an
# attacker could farm authorized-looking activity to suppress a good rule,
# and per-rule FP-rate metrics would be corrupted.
detector = PoisoningDetector()
assert "false_positive" in detector._fp_dispositions
assert "benign_true_positive" not in detector._fp_dispositions