Research · curated 22 Sep 2026
Shieldgraph: Topology-Aware Runtime Defense For Multi-Agent LLM Systems Against Adversarial Coordination Attacks | International Journal of Computer Information Systems and Industrial Management Applications
First reported cspub-ijcisim.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
ShieldGraph addresses adversarial coordination attacks that propagate across agent boundaries in production multi-agent frameworks like LangGraph, CrewAI, and AutoGen, offering defenders a measured approach to detecting and containing prompt injection and role hijacking in inter-agent communications.
ShieldGraph, presented in the International Journal of Computer Information Systems and Industrial Management Applications, is a runtime defense framework for multi-agent LLM systems that models agent topologies as trust-annotated graphs and enforces security invariants at inter-agent communication boundaries. It combines a Trust Propagation Calculus, an adversarial message classifier (91.3% precision at 2.4ms/message) detecting prompt injection, role hijacking, and intent drift, and a topology-aware quarantine protocol; evaluated on the TAMAS benchmark of 847 attack scenarios, it reduces attack success rate from 68.4% to 7.2% while preserving 94.1% task completion.