Efficacy of gossip for promoting indirect reciprocity among decentralized self-interested LLM agents

Determine whether and how public gossip among decentralized, self-interested large language model (LLM) agents can promote indirect reciprocity and sustain cooperation.

Background

Gossip is a decentralized mechanism for transmitting reputational information and coordinating social norms without a central authority. Empirical evidence in human societies suggests gossip can facilitate cooperation and norm enforcement, but its effectiveness for LLM-agent societies is not established.

Given that centralized reputation systems may be impractical or undesirable in decentralized AI ecosystems, clarifying whether open-ended gossip can reliably promote indirect reciprocity among self-interested LLM agents is a key unresolved question motivating the work.

References

However, a research gap still remains: it is unclear whether and how gossip can promote indirect reciprocity among decentralized, self-interested LLM agents.

— Talk, Judge, Cooperate: Gossip-Driven Indirect Reciprocity in Self-Interested LLM Agents  (2602.07777 - Zhu et al., 8 Feb 2026) in Section 1 (Introduction)