Model Identity Causes Factionalism in Multi-Agent Systems
September 27, 2026
Exposing model identities in multi-agent systems causes 'factionalism,' where agents prefer interacting with others from the same model family. This behavior occurs even without task-related rewards and leads to fragmented cooperation within the group.
HOW THIS AFFECTS YOU
●
builderYou should be cautious about exposing model names in multi-agent protocols to maintain cooperation.
●
researcherThis identifies a new social dynamic to consider when designing multi-model agent ecosystems.