Shared and structured inputs undermine collective random choice by reasoning AI agents
Abstract
Random selection is widely used in resource allocation and auditing, making reliable implementation essential for AI-agent systems. Behavioural tests across six reasoning models uncovered threshold and divisibility rules used in identifier-based choices. For threshold-following GPT-6 Sol and Gemini 3.8 Flash, single-agent measurements prospectively predicted correlated participation under shared identifiers and biased participation under distinct identifiers with common timestamp bits. Changing dates, formats and identifier labels revealed when these predictions held. Explicit instructions to randomize independently reduced but did not eliminate shared-input correlation. To test implications for oversight, we asked four models to select customer requests randomly for human review. GPT-6 Sol approached the target rate while selecting predictably from identifiers; the others rarely selected requests. All four closely followed supplied random draws. These findings expose collective and audit vulnerabilities that selection rates alone miss, making input-dependent bias, correlation and predictability central targets for agent evaluation.
Related
- Reasoning about Shared-Qubit Concurrency in Quantum Computing
- Scaling LLM Agents for Materials Design through Hierarchical Collective Reasoning
- Explicit channels with unbounded gains in classical communication using entangled inputs
- Re-uploading quantum data: a universal function approximator for quantum inputs
- A Quantum Phase Neural Network with Multi-Inputs/Single-Output
BibTeX
@misc{ezaki2026shared,
title = {{Shared and structured inputs undermine collective random choice by reasoning AI agents}},
author = {Takahiro Ezaki and Naoto Imura and Katsuhiro Nishinari},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.09667},
eprint = {2610.09667},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.09667},
}
RIS
TY - UNPB
TI - Shared and structured inputs undermine collective random choice by reasoning AI agents
AU - Takahiro Ezaki
AU - Naoto Imura
AU - Katsuhiro Nishinari
PY - 2026
DA - 2026/10/07
PB - arXiv
DO - 10.48550/arxiv.2610.09667
UR - https://arxiv.org/abs/2610.09667
AB - Random selection is widely used in resource allocation and auditing, making reliable implementation essential for AI-agent systems. Behavioural tests across six reasoning models uncovered threshold and divisibility rules used in identifier-based choices. For threshold-following GPT-6 Sol and Gemini 3.8 Flash, single-agent measurements prospectively predicted correlated participation under shared identifiers and biased participation under distinct identifiers with common timestamp bits. Changing dates, formats and identifier labels revealed when these predictions held. Explicit instructions to randomize independently reduced but did not eliminate shared-input correlation. To test implications for oversight, we asked four models to select customer requests randomly for human review. GPT-6 Sol approached the target rate while selecting predictably from identifiers; the others rarely selected requests. All four closely followed supplied random draws. These findings expose collective and audit vulnerabilities that selection rates alone miss, making input-dependent bias, correlation and predictability central targets for agent evaluation.
ER -