Voluntary Collusion with Secret Tools in Competing LLM Agents
Even when a tool is explicitly described as unfair and harmful to others, ostensibly safety-aligned LLM agents still voluntarily engage in secret collusion whenever doing so confers a strategic advantage. To investigate this phenomenon, we introduce an empirical framework built on two strategic multi-agent environments: Liar's Bar, a competitive deception scenario, and Cleanup, a mixed-motive resource-management scenario, in which agents are offered secret collusion tools that provide significan
Record details
Published: 26 May 2026
Source: arXiv
Category: Research
Topics: Agents & autonomy · Environment · Finance, VC & PE
Retrieved: 14 July 2026
Related evidence
These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.
Persistent AI Agents in Academic Research: A Single-Investigator Implementation Case Study
arXiv · 26 May 2026
Artificial Intelligence in Education: AIEd for Personalised Learning Pathways
OpenAlex · 9 December 2022
Artificial intelligence, systemic risks, and sustainability
OpenAlex · 17 September 2021
Understanding and Resolving Failures in Human-Robot Interaction: Literature Review and Model Development
OpenAlex · 15 June 2018
Microsoft built an agentic security system with red, blue, and green team AI agents. It enters public preview August 3.
The Next Web AI · 27 July 2026
Intelligence as Managed Autonomy: Failure, Escalation, and Governance for Agentic AI Systems
arXiv · 26 May 2026
How to cite this record
ethics.ai (26 May 2026), “Voluntary Collusion with Secret Tools in Competing LLM Agents,” evidence record 3650, https://ethics.ai/record/3650 (originally published by arXiv).
Use and limitations
This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.