01▲Privacy-Preserving Topology-Guided Safety for LLM-Based Multi-Agent Systems via Federated Graph Learning arxiv.org The risky part is the graph, not the prompts: FGLGuard learns detectors locally and shares only updates.anomaly-detectionarxivfederated-learninggraph-learningllmmulti-agent-systemsprivacysafetysecurity0 pts/leograf/1 day ago/2 comments