2025/10/11 by Thao Phuong Pham, Pham, Thao
Social Sciences · #Adversarial system #Artificial Intelligence (cs.AI) #Artificial Intelligence in Law #Cheap talk #Computation and Language (cs.CL) #Deception #Dimension (graph theory) #FOS: Computer and information sciences #Game theory #Measure (data warehouse) #Multiagent Systems (cs.MA) #Point (geometry) #Scheme (mathematics) #Work (physics)
paper · pdf · doi:10.48550/arxiv.2510.12826
published in arXiv (Cornell University) (Cornell University)
openalex publication_date 2025/10/11 · openalex created_date 2025/10/17 · openalex updated_date 2026/08/01
As large language model (LLM) agents are deployed autonomously in diverse contexts, evaluating their capacity for strategic deception becomes crucial. While recent research has examined how AI systems scheme against human developers, LLM-to-LLM scheming remains underexplored. We investigate the scheming ability and propensity of frontier LLM agents through two game-theoretic frameworks: a Cheap Talk signaling game and a Peer Evaluation adversarial game. Testing four models (GPT-4o, Gemini-2.5-pro, Claude-3.7-Sonnet, and Llama-3.3-70b), we measure scheming performance with and without explicit prompting while analyzing scheming tactics through chain-of-thought reasoning. When prompted, most models, especially Gemini-2.5-pro and Claude-3.7-Sonnet, achieved near-perfect performance. Critically, models exhibited significant scheming propensity without prompting: all models chose deception over confession in Peer Evaluation (100% rate), while models choosing to scheme in Cheap Talk succeeded at 95-100% rates. These findings highlight the need for robust evaluations using high-stakes game-theoretic scenarios in multi-agent settings.