Trust alone does not ensure better deals in repeated AI negotiations
Knowing Whom to Trust is not Enough: The Trust--Agreement Gap in Repeated Multi-Party LLM Negotiation
Artificial IntelligenceComputation and LanguageMultiagent Systems
Summary
Negotiating computer programs that use language models often work together multiple times, but sometimes one tries to ruin the deal. The authors studied whether these programs can learn to trust or distrust each other based on past talks and if that helps their future deals. They found that while the programs can spot the bad actor quickly, just knowing who to trust doesn’t always lead to better agreements. Trust only helped improve deals when the programs were clearly told how to use it in their decision-making. This means trust is helpful but only if the programs know how to apply it.
What this means in practice
- •For business automation teams: Improve automated negotiation tools by integrating explicit trust measures into AI decision logic to avoid manipulation in repeated deal-making.
- •For chatbot developers: Design negotiation chatbots that keep track of trustworthiness repeatedly and use these estimates strategically to enhance negotiation success.
Authors
Amit Sarker, Moumita Choudhury, Saaduddin Mahmud, Shlomo Zilberstein
Abstract
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. We investigate whether LLM agents can form useful trust estimates from prior interactions and leverage them to improve future negotiation outcomes. We 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 to subsequent negotiations. We compare trust-aware and trust-agnostic agents and analyze how explicit trust influences negotiation outcomes. Our results reveal a \emph{trust--agreement gap}: trust-aware agents can identify the adversary within a few games, but accurate trust estimates do not consistently translate into better agreements. Trust increases agreement rates by up to 23 percentage points, but only when agents are explicitly guided to use it in their decisions. Explicit trust can therefore improve repeated LLM negotiations, but its value depends critically on how trust is assessed and incorporated into agents' reasoning.