Is Monitoring Enough? Strategic Agent Selection For Stealthy Attack in Multi-Agent Discussions

TL;DR

This paper investigates the effectiveness of monitoring in defending against stealthy attacks in multi-agent discussions, revealing that monitoring alone is insufficient to prevent adversarial actions.

Contribution

The study introduces a new attack method tailored for monitored scenarios and demonstrates that effective attacks can still succeed despite continuous monitoring.

Findings

Existing attacks are detectable under monitoring.

Monitoring alone does not fully prevent adversarial attacks.

Effective stealthy attacks remain possible under continuous monitoring.

Abstract

Multi-agent discussions have been widely adopted, motivating growing efforts to develop attacks that expose their vulnerabilities. In this work, we study a practical yet largely unexplored attack scenario, the discussion-monitored scenario, where anomaly detectors continuously monitor inter-agent communications and block detected adversarial messages. Although existing attacks are effective without discussion monitoring, we show that they exhibit detectable patterns and largely fail under such monitoring constraints. But does this imply that monitoring alone is sufficient to secure multi-agent discussions? To answer this question, we develop a novel attack method explicitly tailored to the discussion-monitored scenario. Extensive experiments demonstrate that effective attacks remain possible even under continuous monitoring, indicating that monitoring alone does not eliminate adversarial risks.