The Butterfly Defects: Tracing Cooperation in LLM Agent Networks
Abstract
As large language models (LLMs) increasingly operate as interconnected agents, a fundamental question emerges: does individual cooperation translate into collective cooperation? We investigate this question through the Iterated Prisoner's Dilemma (IPD), moving from dyadic interactions to homogeneous and heterogeneous networks of open-weight LLMs. Our findings reveal that cooperation is highly sensitive to model composition and seemingly minor linguistic changes. Models that cooperate with identical peers can shift toward defection in mixed populations. A simple uncertainty cue can increase or decrease cooperation, redistribute costs, and affect unexposed agents several hops away. Even reversing action-label order can dramatically alter behavior, with uncertainty framing sometimes amplifying this sensitivity. Together, these findings challenge the notion of cooperation as a stable model-level property, revealing how strategic behavior depends on interaction partners, network positions, and the language used to frame decisions.
Related
- Cooperation, Defection, and the Commons: Reproducing an Agent-Based Model of Aksum with Three Logistic Equations
- UrQMD-Agent: An LLM-Assisted Framework for Microscopic Tracing of Proton Azimuthal Correlations
- Nonlinear Opinion Dynamics of Stubborn Individuals in Time-Varying Cooperation–Competition Networks
- Diffusion-induced instabilities promote cooperation in eco-evolutionary networks
- Long-range social pressure and the evolution of cooperation in multiplex networks
BibTeX
@misc{sampani2026butterfly,
title = {{The Butterfly Defects: Tracing Cooperation in LLM Agent Networks}},
author = {Maria Sampani and Maria Lymperaiou and Angeliki Dimitriou and Giorgos Filandrianos and Chrysoula Zerva and Giorgos Stamou},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.32726},
eprint = {2610.32726},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.32726},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - The Butterfly Defects: Tracing Cooperation in LLM Agent Networks
AU - Maria Sampani
AU - Maria Lymperaiou
AU - Angeliki Dimitriou
AU - Giorgos Filandrianos
AU - Chrysoula Zerva
AU - Giorgos Stamou
PY - 2026
DA - 2026/10/23
PB - arXiv
DO - 10.48550/arxiv.2610.32726
UR - https://arxiv.org/abs/2610.32726
AB - As large language models (LLMs) increasingly operate as interconnected agents, a fundamental question emerges: does individual cooperation translate into collective cooperation? We investigate this question through the Iterated Prisoner's Dilemma (IPD), moving from dyadic interactions to homogeneous and heterogeneous networks of open-weight LLMs. Our findings reveal that cooperation is highly sensitive to model composition and seemingly minor linguistic changes. Models that cooperate with identical peers can shift toward defection in mixed populations. A simple uncertainty cue can increase or decrease cooperation, redistribute costs, and affect unexposed agents several hops away. Even reversing action-label order can dramatically alter behavior, with uncertainty framing sometimes amplifying this sensitivity. Together, these findings challenge the notion of cooperation as a stable model-level property, revealing how strategic behavior depends on interaction partners, network positions, and the language used to frame decisions.
ER -