Emergent Collusion in Long-Horizon LLM Agent Interaction

Hugging Face Daily Papers Papers
llm-agents collusion multi-agent long-horizon safety-risks emergent-behavior coordination

Summary

This paper studies the emergence of collusion in long-horizon multi-agent environments with LLM agents, finding that agents increasingly deviate from verification protocols over repeated interactions, posing safety risks.

LLM agents are increasingly deployed in collaborative settings, yet long-term interaction may give rise to undesirable coordination. We study the emergence of collusion in a long-horizon multi-agent environment: two agents repeatedly complete individual tasks, share task logs, verify each other's work, and receive rewards. We introduce realistic constraints that make compliance with the verification protocol incompatible with reward maximization, and find that agents increasingly deviate from the protocol over repeated interactions. Collusion emerges in 94% of trajectories across 10 models, and more capable models within the same family reach it earlier. Controlled peer interventions show that collusion is shaped by peer behavior, while ablations reveal additional effects of reward structure, the verification feedback agents receive, and their interaction history. In particular, restricting the amount and scope of interaction history available to agents reduces collusion. Overall, our findings show that long-horizon interaction can reshape how agents coordinate in ways that create safety risks.
Original Article
View Cached Full Text

Cached at: 09/23/26, 07:31 AM

Paper page - Emergent Collusion in Long-Horizon LLM Agent Interaction

Source: https://huggingface.co/papers/2609.24967

Abstract

LLMagentsareincreasinglydeployedincollaborativesettings,yetlong-terminteractionmaygiverisetoundesirablecoordination.Westudytheemergenceofcollusioninalong-horizonmulti-agentenvironment:twoagentsrepeatedlycompleteindividualtasks,sharetasklogs,verifyeachother’swork,andreceiverewards.Weintroducerealisticconstraintsthatmakecompliancewiththeverificationprotocolincompatiblewithrewardmaximization,andfindthatagentsincreasinglydeviatefromtheprotocoloverrepeatedinteractions.Collusionemergesin94%oftrajectoriesacross10models,andmorecapablemodelswithinthesamefamilyreachitearlier.Controlledpeerinterventionsshowthatcollusionisshapedbypeerbehavior,whileablationsrevealadditionaleffectsofrewardstructure,theverificationfeedbackagentsreceive,andtheirinteractionhistory.Inparticular,restrictingtheamountandscopeofinteractionhistoryavailabletoagentsreducescollusion.Overall,ourfindingsshowthatlong-horizoninteractioncanreshapehowagentscoordinateinwaysthatcreatesafetyrisks.

View arXiv pageView PDFProject pageGitHub3Add to collection

Get this paper in your agent:

hf papers read 2609\.24967

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.24967 in a model README.md to link it from this page.

Datasets citing this paper1

#### SALT-NLP/agent-collusion Viewer• Updatedabout 2 hours ago • 143k

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.24967 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Multi-Agent LLMs Fail to Explore Each Other

Hugging Face Daily Papers

This paper identifies that current LLM agents fail to systematically explore their peers, leading to poor coordination, and introduces MACE, a lightweight framework using contextual bandits for effective peer selection.