Search the Atlas

Search risks, controls, and glossary terms

HighAgenticAI Lateral MovementDemonstrated

Cross-System Agent Traversal

Multi-Agent Security

Description

Compromised agents use legitimate access and trust relationships to traverse connected systems. Agents acting as insider threats leverage authorised access to reach unauthorised resources across system boundaries.

Example scenario

Compromised customer service agent uses its database access to enumerate internal APIs and reach HR system data.

Real-world evidenceDemonstrated

Academic studies and structured red-team evaluations have confirmed that LLMs produce materially different outputs for semantically equivalent prompts across runs, but there is no widely reported production incident where this inconsistency caused a discrete, documented harm event at scale in a deployed financial application.

Primary mitigations

  • Strict agent network segmentation
  • cross-system access monitoring
  • lateral movement detection
  • agent access path analysis.

Detection signals

Lateral movement detection events; cross-boundary access anomalies; agent authentication failures.

Mitigating controls

5
Dual coverage

Related risks in Multi-Agent Security