Sharing is caring: data sharing in multi-agent supply chains

A new research paper from Wan Wang, Haiyan Wang, and Adam Sobey tackles a critical realism gap in multi-agent AI systems for supply chain optimization. Published on arXiv (2602.24074), 'Sharing is caring: data sharing in multi-agent supply chains' challenges the common assumption that AI agents in supply networks have full system observability through shared data. The researchers argue this is unrealistic, as companies are often reluctant to share proprietary information. Instead, they propose a more practical Hidden-Markov Process model where a factory agent can strategically choose what information to share downstream—including the option to lie.

The study's key finding is that optimal data-sharing strategy depends entirely on demand conditions. In low-demand scenarios, truth-telling provides the most significant benefits across all supply chain actors. However, in high-demand environments where flexibility is limited, strategic lying by the factory agent can produce marginal overall system improvements, though it primarily benefits the factory itself. The research demonstrates that combining these strategic communication approaches with cooperative reward shaping—where agents receive rewards for helping others—can boost overall system performance. This work provides a crucial framework for developing more realistic, privacy-preserving multi-agent systems that don't require full data transparency between potentially competing business entities.