Evo-MARL: Co-Evolutionary Multi-Agent Reinforcement Learning for Internalized Safety

TL;DR

Evo-MARL introduces a co-evolutionary multi-agent reinforcement learning framework that internalizes safety, enabling agents to jointly develop defensive capabilities against adversarial threats without external modules.

Contribution

It presents a novel co-evolutionary MARL approach that internalizes safety mechanisms within agents, reducing reliance on external safety modules and enhancing robustness.

Findings

Reduces attack success rates by up to 22%.

Increases reasoning accuracy by up to 5%.

Demonstrates improved safety and utility in multi-agent systems.

Abstract

Multi-agent systems (MAS) built on multimodal large language models exhibit strong collaboration and performance. However, their growing openness and interaction complexity pose serious risks, notably jailbreak and adversarial attacks. Existing defenses typically rely on external guard modules, such as dedicated safety agents, to handle unsafe behaviors. Unfortunately, this paradigm faces two challenges: (1) standalone agents offer limited protection, and (2) their independence leads to single-point failure-if compromised, system-wide safety collapses. Naively increasing the number of guard agents further raises cost and complexity. To address these challenges, we propose Evo-MARL, a novel multi-agent reinforcement learning (MARL) framework that enables all task agents to jointly acquire defensive capabilities. Rather than relying on external safety modules, Evo-MARL trains each agent to simultaneously perform its primary function and resist adversarial threats, ensuring robustness without increasing system overhead or single-node failure. Furthermore, Evo-MARL integrates evolutionary search with parameter-sharing reinforcement learning to co-evolve attackers and defenders. This adversarial training paradigm internalizes safety mechanisms and continually enhances MAS performance under co-evolving threats. Experiments show that Evo-MARL reduces attack success rates by up to 22% while boosting accuracy by up to 5% on reasoning tasks-demonstrating that safety and utility can be jointly improved.