Can LLM Agents Simulate Dynamic Networks? A Case Study on Email Networks with Phishing Synthesis

This collection highlights emergent security and misinformation risks from recent network/LLM research. Key findings: (1) LLM multi-agent systems can be extended (data-driven event triggers + Hawkes temporal models) to synthesize realistic, long-horizon phishing campaigns and reproduce macroscopic network dynamics (paper + code: Graph-COM/NSL). (2) GraphMind shows LLM-driven bots can be made graph-aware to construct human-like social networks (GraphMind-Botnet), substantially degrading both text- and graph-based bot-detection models. (3) Misinformation and health-harm risks are exemplified by: