# Shared and structured inputs undermine collective random choice by reasoning AI agents

arXiv: https://arxiv.org/abs/2610.09667 · HTML: https://mlsanity.com/paper/2610.09667/

## Key facts

- No citing papers in the map yet (data through 7 Oct 2026).
- Placed in Social simulation and multi-agent debate, one of 16 directions in LLM agents and tool use.
- Builds on 11 of its 37 references that are in the map.
- First posted on arXiv 7 Oct 2026.

## Placed in

- LLM agents and tool use / [Social simulation and multi-agent debate](https://mlsanity.com/agents/social-simulation-and-multi-agent-debate/)

## Builds on (most cited first, 11 of 11)

- [Reasoning Models Don't Always Say What They Think](https://mlsanity.com/paper/2505.05410/) (2025)
- [Heads, Tails, and AI Fails: LLMs, Randomness, and Human Judgments](https://mlsanity.com/paper/2406.00092/) (2024)
- [Subversion Strategy Eval: Can language models statelessly strategize to subvert control protocols?](https://mlsanity.com/paper/2412.12480/) (2024)
- [Playing games with Large language models: Randomness and strategy](https://mlsanity.com/paper/2503.02582/) (2025)
- [Emergent Social Dynamics of LLM Agents in the El Farol Bar Problem](https://mlsanity.com/paper/2509.04537/) (2025)
- [The Illusion of Stochasticity in LLMs](https://mlsanity.com/paper/2604.06543/) (2026)
- [Strategic Algorithmic Monoculture: Experimental Evidence from Coordination Games](https://mlsanity.com/paper/2604.09502/) (2026)
- [Tacit Coordination of Large Language Models](https://mlsanity.com/paper/2601.22184/) (2026)
- [Subversion via Focal Points: Investigating Collusion in LLM Monitoring](https://mlsanity.com/paper/2507.03010/) (2025)
- [Frontier Models Can Take Actions at Low Probabilities](https://mlsanity.com/paper/2603.02202/) (2026)
- [Do LLMs Beat Nash? Testing Decentralized Coordination in Self-Play Multi-Agent Games](https://mlsanity.com/paper/2608.12547/) (2026)

## Abstract (arXiv)

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.