When Collaboration Becomes a Trigger: Collective Evidence-Threshold Backdoors in Multi-Agent Systems

Abstract

LLM-based multi-agent systems (MAS) extend LLM capabilities through iterative communication and shared contexts. However, this collaboration introduces a vulnerability: backdoor behavior can be activated when peer evidence reaches a hidden threshold, rather than being determined by any single message. We introduce a collective evidence-threshold backdoor paradigm for MAS and Boundary-Conditioned Backdoor Injection (BCBI), which constructs counterfactual boundary pairs to separate benign behavior

Categories

Framework mappings

Suggested from the entry's categories.

@misc{xiao2026when,
  title = {{When Collaboration Becomes a Trigger: Collective Evidence-Threshold Backdoors in Multi-Agent Systems}},
  author = {Jiahao Xiao and Lei Feng and Min-Ling Zhang},
  year = {2026},
  month = aug,
  eprint = {2608.01085},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/721e0bc978010a3cb051e3bc43537d30dddcc811}
}