The field of multi-agent systems is moving towards developing more robust and scalable communication strategies, with a focus on engineered approaches that can handle complex and dynamic tasks. Researchers are exploring new methods to incentivize agents to follow specific strategies, such as using optimal messaging strategies and adaptive tampering frameworks. Additionally, there is a growing interest in internalizing safety mechanisms within multi-agent systems, rather than relying on external guard modules. This approach enables agents to jointly acquire defensive capabilities, ensuring robustness without increasing system overhead. Notable papers in this area include:

• Engineered over Emergent Communication in MARL for Scalable and Sample-Efficient Cooperative Task Allocation in a Partially Observable Grid, which demonstrates the superiority of engineered communication strategies in cooperative tasks.
• Evo-MARL: Co-Evolutionary Multi-Agent Reinforcement Learning for Internalized Safety, which proposes a novel framework for training agents to simultaneously perform their primary function and resist adversarial threats.