AgentCollusionAI security startupResearch
Our open research agenda.
Papers, technical notes, and empirical work on hidden coordination and security in multi-agent systems.
Published work
Latest outputs
Detecting AI Collusion in Multi-Agent Systems
A systems and game-theoretic framework using temporal graphs, counterfactuals, incentives, and adaptive evaluation.
Read research note ExplainerWhat Is AI Collusion?
A precise introduction to cooperation, covert coordination, and why single-agent safety checks miss the relationship.
Read explainer Visual guide / 日本語Agent CollusionとA2Aを、8枚で理解する
Agent Collusion、A2A、エージェント結託・共謀、A2A取引を一般向けの図で整理します。
ビジュアルガイドを見るResearch scope
We study behavior that only becomes visible when agents, incentives, messages, tools, and outcomes are analyzed together.
Open research agenda
How do we tell cooperation from collusion?
- 01
How can we distinguish beneficial cooperation from covert coordination?
- 02
How should we evaluate possible signals against adaptive, strategic agents?
- 03
What responses could reduce risk without destroying useful autonomy?
- 04
How do capability, speaking order, repetition, memory access, and tool authority let one agent dominate a group?
