<ul><li>Researchers have developed an adversarial attack that can bypass safety mechanisms in multi-agent Large Language Model (LLM) systems.</li><li>The attack optimizes prompt distribution across latency and bandwidth-constrained network topologies to maximize attack success rate while minimizing detection risk.</li><li>The method outperforms conventional attacks, exposing critical vulnerabilities in multi-agent systems.</li><li>Existing defenses, including variants of Llama-Guard and PromptGuard, fail to prohibit the attack.</li></ul>

$\textit{Agents Under Siege}$: Breaking Pragmatic Multi-Agent LLM Systems with Optimized Prompt Attacks

Discover more