Finding the Weakest Link: Adversarial Attack against Multi-Agent Communications

2026-05-13

AI Summary

This work addresses the vulnerability of multi-agent systems to adversarial attacks under communication perturbations, a challenge inadequately tackled by existing methods that struggle to pinpoint critical weaknesses. The authors propose a Jacobian-based gradient-driven message selection strategy coupled with a victim localization mechanism to precisely identify the messages, agents, and timesteps that most significantly degrade system performance. Additionally, two novel adversarial loss functions are introduced to balance attack success rate against impact severity. Experimental evaluation across diverse environments—including Navigation, PredatorPrey, and TrafficJunction—demonstrates the efficacy of the proposed approach: it consistently matches or outperforms random selection and substantially enhances attack effectiveness in half of the 30 tested scenarios.

Abstract

Multi-agent systems rely on communication for information sharing and action coordination, which exposes a vulnerability to attacks. We investigate single-victim communication perturbation attacks against Multi-Agent Reinforcement Learning-trained systems and propose methods that use gradient information from the Jacobian to identify which messages, agent, and timesteps are most susceptible to attack and have the greatest impact on the system. We enhance these methods with two proposed adversarial loss functions that trade-off attack success for attack impact which also create more effective perturbations. We empirically demonstrate the effectiveness of our methods against two different multi-agent communication methods in navigation, PredatorPrey, and TrafficJunction environments. Our results show that our novel message selection method achieves a similar or greater impact than random message selection across almost all tested scenarios. Our victim selection, message selection, tempo, and loss functions improve attack effectiveness in half of the thirty scenarios we tested.

Problem

• adversarial attack
• multi-agent communication
• vulnerability
• message perturbation
• attack impact

Innovation

• adversarial attack
• multi-agent communication
• Jacobian gradient
• message selection
• adversarial loss function