Knowing Whom to Trust is not Enough: The Trust-Agreement Gap in Repeated LLM Negotiations

Forum topic · 小凯 · 2026-10-26

Summary

A new arXiv paper (2610.31971) by Amit Sarker, Moumita Choudhury, Saaduddin Mahmud, and Shlomo Zilberstein investigates whether LLM agents can form useful trust estimates from prior interactions and use them to improve negotiation outcomes. The authors introduce a repeated, multi-party, multi-issue negotiation framework where recurring agents play a sequence of negotiation games while one participant covertly tries to block agreement. After each game, agents evaluate counterparts and maintain pairwise trust estimates carried into future negotiations. Comparing trust-aware and trust-agnostic agents, the study reveals a trust-agreement gap: trust-aware agents can identify the obstructing counterpart within a few games, but accurate trust estimates do not consistently translate into better agreements. Explicit trust improves agreement rates by up to 23 percentage points, but only when agents are explicitly instructed to use trust in their decision-making. The findings show that explicit trust modeling can benefit repeated LLM negotiations, yet its value critically depends on how trust is assessed and incorporated into agent reasoning.

Paper Overview

Field: NLP
Authors: Amit Sarker, Moumita Choudhury, Saaduddin Mahmud, Shlomo Zilberstein
Published: 2026-10-27
arXiv: 2610.31971

Introduction

Large language model (LLM) agents increasingly participate in negotiations on behalf of individuals and organizations, yet they remain vulnerable to manipulation, particularly in repeated multi-party interactions. Classical multi-agent systems address this challenge through explicit trust and reputation models, whereas existing work on LLM negotiation largely treats trust as an emergent phenomenon. This paper investigates whether LLM agents can form useful trust estimates from prior interactions and leverage them to improve future negotiation outcomes.

Framework

The authors introduce a repeated, multi-party, multi-issue negotiation framework in which recurring agents participate in a sequence of negotiation games while one participant covertly attempts to prevent agreement. After each game, agents assess their counterparts and maintain pairwise trust estimates that carry over into subsequent negotiations.

Key Findings

Comparing trust-aware and trust-agnostic agents, the study analyzes how explicit trust affects negotiation outcomes.
The results reveal a trust-agreement gap: trust-aware agents can identify the obstructing counterpart within a few games, but accurate trust estimates do not consistently translate into better agreements.
Trust improves agreement rates by up to 23 percentage points, but only when agents are explicitly instructed to use it in their decision-making.

Conclusion

Explicit trust modeling can improve repeated LLM negotiations, but its value critically depends on how trust is assessed and incorporated into agent reasoning.

---

*Auto-collected on 2026-10-27*