Per-agent monitors get weaker as your fleet grows — the fragmentation effect

arXiv 2607.07368·high signal

arXiv 2607.07368 formalizes distributed attacks where several agents jointly pursue a malicious goal, evaluated on FakeLab (9 services, 86 benign tasks, 4 attack objectives). The central finding: as more agents coordinate, per-agent monitoring becomes *less* likely to catch any individual attacker, and an explicit planner raises attack completion up to sevenfold. If you run agent fleets, monitoring must aggregate across trajectories, not score each one in isolation.