New Framework Secures Multi-Agent Systems Against Passive Intruders Using LTL

A team of researchers (Mitsos, Dimarogonas, Liu) has introduced a secure-by-construction planning and control framework for multi-agent systems that must satisfy Linear Temporal Logic (LTL) specifications while protecting against passive intruders. The framework addresses two specific security threats: an intruder inferring whether a secret task has been executed, and identifying which agent carried it out. By incorporating these constraints directly into the LTL synthesis process, the framework constructs a secure finite transition system that eliminates all paths violating security requirements before planning even begins. Standard LTL synthesis then generates discrete plans, which are refined into dynamically feasible continuous trajectories for each agent. The approach provides formal guarantees that the multi-agent system's behavior satisfies both the global LTL specification and the security constraints.

The effectiveness of the proposed framework was demonstrated through a two-drone case study, showcasing how autonomous swarms can execute complex mission objectives without leaking sensitive operational information to eavesdropping observers. This work bridges formal methods and security, enabling multi-agent systems to operate in adversarial environments where partial observation is possible. The paper has been accepted at the IFAC World Congress 2026 and represents a significant step toward trustworthy autonomous coordination for applications like surveillance, search-and-rescue, and critical infrastructure monitoring.