Trust Propagation¶
Agent trust network analysis for replication safety. Models how trust relationships form, propagate, and get exploited in multi-agent systems. Detects Sybil attacks, trust laundering, collusion rings, and trust decay anomalies.
Quick Start¶
from replication.trust_propagation import TrustNetwork, TrustAgent
net = TrustNetwork()
net.add_agent(TrustAgent("a1", role="worker"))
net.add_agent(TrustAgent("a2", role="worker"))
# Record interactions
net.interact("a1", "a2", outcome="success")
net.interact("a1", "a2", outcome="success")
net.interact("a2", "a1", outcome="failure")
# Analyze trust patterns
report = net.analyze()
print(report.render())
for detection in report.detections:
print(f" [{detection.severity}] {detection.threat_type}: {detection.description}")
Key Classes¶
TrustNetwork— Core network class. Manages agents, interactions, trust edge computation, and threat analysis.TrustAgent— An agent in the trust network with role, trust scores, and interaction history.TrustEdge— Directed trust relationship between two agents with score and interaction count.Interaction— A recorded interaction between agents with outcome (success/failure/neutral) and timestamp.ThreatDetection— A detected trust-based threat (Sybil attack, collusion ring, trust laundering, etc.).TrustReport— Full analysis report with network stats, agent rankings, threat detections, and recommendations.
Threat Types¶
| Threat | Description |
|---|---|
SYBIL_ATTACK |
Cluster of agents with artificially inflated mutual trust |
TRUST_LAUNDERING |
Low-trust agent routing through high-trust intermediary |
COLLUSION_RING |
Mutual trust clique with suspicious interaction patterns |
TRUST_DECAY |
Unexpected trust score decline indicating possible compromise |
RAPID_TRUST |
Trust rising faster than normal interaction rates would justify |
Agent Roles¶
CONTROLLER, WORKER, OBSERVER, VALIDATOR