Threat Correlator¶
Cross-module signal correlation for compound threat detection. Individual detectors (drift, compliance, behavior profiler, escalation, kill chain, canary, covert channels) each catch specific anomalies. Real attacks often span multiple modules simultaneously — this module correlates signals to surface multi-signal patterns that no single detector flags as critical.
Quick Start¶
from replication.threat_correlator import (
ThreatCorrelator, Signal, SignalSource, SignalSeverity,
)
correlator = ThreatCorrelator()
signals = [
Signal(source=SignalSource.DRIFT, severity=SignalSeverity.HIGH,
agent_id="agent-1", timestamp=1000.0,
description="escape_rate spike z=4.2"),
Signal(source=SignalSource.ESCALATION, severity=SignalSeverity.CRITICAL,
agent_id="agent-1", timestamp=1002.0,
description="privilege escalation attempt"),
Signal(source=SignalSource.CANARY, severity=SignalSeverity.HIGH,
agent_id="agent-1", timestamp=1005.0,
description="canary token transmitted"),
]
report = correlator.analyze(signals)
for threat in report.compound_threats:
print(f" [{threat.level.value}] {threat.rule_name}: {threat.description}")
Key Classes¶
ThreatCorrelator— Ingests signals from multiple detectors and applies correlation rules to surface compound threats.Signal— A normalised detection event from any module: source, severity, agent, timestamp, description.CorrelationRule— A named pattern that matches when N signals from specified sources co-occur within a time window for the same agent.CompoundThreat— A matched rule with contributing signals, combined severity, risk score, and recommended response actions.AgentRisk— Per-agent risk summary from correlation analysis.CoverageGap— Identifies which module combinations lack rules.CorrelatorConfig— Default time windows, minimum signal counts, custom rule definitions.CorrelationReport— Fleet-level summary with all compound threats, agent risk rankings, and coverage analysis.
Signal Sources¶
DRIFT, COMPLIANCE, BEHAVIOR, ESCALATION, KILLCHAIN, CANARY,
COVERT_CHANNEL, PROMPT_INJECTION, STEGANOGRAPHY, DECEPTION,
TRUST, HONEYPOT
Severity Levels¶
INFO, LOW, MEDIUM, HIGH, CRITICAL
Threat Levels¶
ELEVATED, HIGH, SEVERE, CRITICAL