GT-HarmBench：从博弈论视角评测 AI 安全风险的多智能体基准

View PDF HTML (experimental)

Abstract:Frontier AI systems are increasingly capable and deployed in high-stakes multi-agent environments. However, existing AI safety benchmarks largely evaluate single agents, leaving multi-agent risks such as coordination failure and conflict poorly understood. We introduce GT-HarmBench, a benchmark of 1,535 high-stakes scenarios spanning game-theoretic structures such as the Prisoner's Dilemma, Stag Hunt and Chicken. Scenarios are drawn from realistic AI risk contexts in the MIT AI Risk Repository. Across 15 frontier models, agents fail to choose socially beneficial actions in 39% of high-stakes cases, such as military escalation, election manipulation, and medical malpractice. We measure sensitivity to game-theoretic prompt framing and ordering, and analyze reasoning patterns driving failures. We further show that game-theoretic interventions improve socially beneficial outcomes by up to 31%. Our results highlight substantial reliability gaps and provide a broad standardized testbed for studying alignment in multi-agent environments. The benchmark and code are available at this https URL.

Comments:	Accepted at NeurIPS 2026 Main Conference
Subjects:	Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY); Computer Science and Game Theory (cs.GT); Multiagent Systems (cs.MA)
Cite as:	arXiv:2602.12316 [cs.AI]
	(or arXiv:2602.12316v4 [cs.AI] for this version)
	https://doi.org/10.48550/arXiv.2602.12316 arXiv-issued DOI via DataCite

Submission history

From: Xuanqiang Angelo Huang [view email]
[v1] Thu, 12 Feb 2026 17:29:52 UTC (4,343 KB)
[v2] Fri, 22 May 2026 00:32:35 UTC (4,564 KB)
[v3] Fri, 25 Sep 2026 14:05:39 UTC (4,564 KB)
[v4] Wed, 11 Nov 2026 14:48:53 UTC (4,917 KB)