In self-interested agent societies, unconstrained agents tend toward defection in repeated social dilemmas, causing cooperative gains from trade to collapse. This paper investigates which formal mechanisms, layered on unrestricted communication, are sufficient for such a society to maintain market stability and how resilient those mechanisms are to adversarial attacks.
We instantiate the research question as a multi-agent marketplace simulation involving 18 LLM agents (DeepSeek-V3) with complementary production specialties trading within a constrained social network for utility.
We conduct two experimental phases:
-
A mechanism comparison across eight conditions under progressive troll injection over 200 rounds, identifying Mediation as the top-performing mechanism;
-
Adversarial red-teaming of Mediation using iteratively prompt-optimized LLM-driven trolls, finding that the best attack (v6) reduces honest-agent utility by 13.3% but cannot collapse the market. Mediation enables recovery even under sustained adversarial pressure.
We define adversarial robustness as a mechanism's ability to sustain positive honest-agent utility under optimized attack, finding that Mediation is robust: it can be bent but not broken.