About Cooperative Behavior of LLM Agents in Different Repeated Social Dilemma Game Contexts

Abstract

This research investigates potential variations in cooperative behavior among Large Language Model agents in a base environment and six context-augmented Prisoner's Dilemma and Snowdrift game environments. We observe a positive impact of context augmentation on cooperative behavior, but not consistent increases in payoff. Also, we identify context-augmented exit-from-Nash-equilibrium and opponent alignment behaviors only under restricted experimental setups. Finally, the results are discussed in light of potential threats to the validity of the research. © 2026 IEEE.