AI 沙盒：威胁模型、分类法与测量框架
AI Sandboxes: A Threat Model, Taxonomy, and Measurement Framework
AI 导读
arXiv 论文提出 AI 沙盒的测量框架，界定沙盒装置、五类重叠原型和 15 个相对声明（claim-relative）的测量维度，并用最弱环节许可规则限定部署声明。该规则与威胁模型相连，覆盖对外危害与评估装置本身的腐化，并通过驾驶、工业控制、工具使用智能体和视觉-语言-动作操作四个实例演示，同时把所需工件映射到标准与监管文件。
正文 · 原文
Abstract:AI sandboxes produce evidence about systems that may act through software tools, networked devices, or physical actuators. The interpretation of that evidence depends on the boundary of the experiment. We define a sandbox apparatus, distinguish five overlapping archetypes, and specify fifteen claim-relative measurement dimensions. A weakest-link licensing rule caps a deployment claim at its weakest necessary evidence dimension. We connect this rule to a threat model covering both outward harm and corruption of the evaluation apparatus, then apply it to four worked examples spanning driving, industrial control, tool-using agents, and vision-language-action manipulation. We also map the required artifacts to standards and regulatory instruments. The framework identifies which claims a bounded experiment supports, which failures its controls contain, and which assumptions require further validation before deployment.
| Comments: | 42 pages, 8 figures, 7 tables |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Robotics (cs.RO); Software Engineering (cs.SE) |
| Cite as: | arXiv:2606.18532 [cs.CR] |
| (or arXiv:2606.18532v2 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2606.18532 arXiv-issued DOI via DataCite |
Submission history
From: Inderjeet Singh [view email]
[v1]
Tue, 16 Jun 2026 22:57:24 UTC (1,985 KB)
[v2]
Tue, 10 Nov 2026 22:45:06 UTC (1,996 KB)
来源：arXiv cs.AI · arxiv.org