Behavioural Signatures of Risk-Sensitive Decision-Making in Large Language Models

2026-07-11

AI Summary

This study investigates whether large language models (LLMs) exhibit stable and interpretable risk-taking behaviors under uncertainty. By constructing a multi-agent interaction framework based on no-limit Texas Hold’em poker, the work integrates homogeneous self-play and heterogeneous mixed-strategy competitions, introducing quantitative metrics such as engagement and proactiveness to systematically analyze model behavior in controlled risk scenarios. The research reveals, for the first time, that LLMs possess individual-specific yet robust risk preference spectra—ranging from conservative to aggressive—that remain behaviorally robust despite changes in opponents. Furthermore, under global pressure and resource constraints, these models demonstrate diverse adaptive strategies, highlighting significant heterogeneity in their contextual adaptability.

Abstract

As large language models (LLMs) are increasingly used in decision support, it is important to understand whether their choices under uncertainty exhibit stable and interpretable behavioural regularities. Human decision-making combines relatively persistent risk preferences with context-dependent adjustment, yet it remains unclear whether analogous behavioural structure can be observed in LLM-based decision systems. Here we examine this question using a controlled multi-model framework based on no-limit Texas Hold'em, where behaviour is quantified by Participation, measuring voluntary engagement in uncertain opportunities, and Proactiveness, measuring pre-flop risk escalation. Across homogeneous self-play and heterogeneous mixed-model interactions, frontier LLMs exhibit stable, model-specific risk profiles, forming a spectrum from conservative to aggressive decision styles. These profiles remain largely robust under changing opponent composition, while the most conservative and most aggressive models diverge further in mixed settings. Under global risk pressure and personal resource constraint, models adapt in structured but heterogeneous ways, ranging from broad behavioural contraction to selective de-escalation and near-invariant behaviour. These findings suggest that LLMs differ not only in baseline risk disposition, but also in the risk signals they respond to and the flexibility with which they adjust, providing a behavioural basis for auditing risk-sensitive decision-making in interactive settings. Our code is publicly available at: https://github.com/XuankunRong/AgentTexasPoker.

Problem

• risk-sensitive decision-making
• large language models
• behavioural regularities
• uncertainty
• decision support

Innovation

• risk-sensitive decision-making
• large language models
• behavioural signatures
• multi-agent interaction
• uncertainty quantification