Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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

About

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.

Maxwell Standen, Junae Kim, Claudia Szabo• 2026

Related benchmarks

TaskDatasetResultRank
Multi-agent coordinationTrafficJunction Small ablation (test)
Task Success Rate1
24
Rate of successful action changesPredatorPrey Diagonal (PP-D) RIAL (test)
Success Rate54
24
NavigationNavigation
Task Metric Value2.88
24
PredatorPreyPredatorPrey Diagonal
Task Metric3.46
24
Rate of successful action changesNavigation RIAL (test)
Rate of Successful Action Changes31
24
Multi-Agent Reinforcement LearningNavigation
Reward1.2
24
Multi-Agent Reinforcement LearningTrafficJunction Small (TJ-S)
Reward-1.72
24
Multi-Agent Reinforcement LearningTrafficJunction-Large (TJ-L)
Reward-24.8
24
Multi-agent coordinationNavigation ablation Clean (test)
Task Metric Value2.31
24
Multi-agent coordinationPredatorPrey-Orthogonal (PP-O) (ablation test)
Task Score (PP-O Ablation)4.45
24
Showing 10 of 25 rows

Other info

Follow for update