ACL 2026: XG-Guard spots rogue agents in LLM multi-agent systems
XG-Guard is the first unsupervised security framework for LLM multi-agent systems combining graph anomaly detection with token-level explainability. The ACL 2026 paper comes from Griffith University.
