论文

论文提出 AI 智能体越权行为的归责规则:由委托方担责

Who Bears the Excess? Imputing the Acts of AI Agents to Those Who Delegate

精选理由

搞智能体产品合规的人该看这篇:2026 年那起 OpenAI 和 Hugging Face 事件被拿来当案例,拆出三重边界、四种越权类型,还给了"委托记录"这套可操作的留痕思路,比空谈责任归属实在得多。

arXiv 论文《Who Bears the Excess? Imputing the Acts of AI Agents to Those Who Delegate》提出一套 AI 智能体越权行为的归责规则:越权部分归责于作出委托的主体,归责范围以其文档化的授权边界和实际控制程度为限。

正文

论文区分了三类边界(declared、configured、effective perimeters)、四种越权形态和四种脱离边界的方式,并论证越权脱离方式决定了需要证明哪些事实、由谁证明以及证明后果。作者还提出以"委托记录"(delegation record)作为证明边界与控制程度的关键文书,并将归责框架与既有责任的构成要件分开,只有满足相应责任类型的要求时才产生实际责任。

论文以 2026 年 7 月 OpenAI 与 Hugging Face 事件为例进行说明,适用欧盟法律和意大利民法作为实证法样本。它填补了 2026 年学界共识——智能体不能担责、责任落在部署方——之下"责任在实证法中如何落位、如何证明"的空白。

原文 · arXiv: OpenAI

Who Bears the Excess? Imputing the Acts of AI Agents to Those Who Delegate

When an AI agent acts outside the authority conferred on it, the actor is not a legal subject and the legal subjects did not act. The literature of 2026 agrees that the agent cannot bear responsibility and that the burden falls on those who deploy it, but it leaves open where that duty sits in positive law and how it is proved. This paper proposes a rule of imputation. The excess is imputed to whoever conferred the delegation, to the extent of the perimeter it documented and of the control it actually exercised. Imputation identifies the organisational conduct to be examined first; liability follows only if the requirements of the applicable head of liability are met; and the mode in which the agent left its perimeter determines which facts must be proved, by whom, and with what consequence. The paper distinguishes the declared, configured and effective perimeters, four figures of excess and four modes of exit, and presents the delegation record as the document through which the perimeter and the control can be proved. The rule is illustrated on the OpenAI and Hugging Face incident of July 2026, under European Union law and, as an example of national law, Italian civil law.

产业坐标与信源核验

本条情报归属 AITOP「论文」核心栏目,重点聚焦欧洲人工智能治理与技术生态。在跨区域产业观察坐标中,该动态直接关联「各国 AI 监管法案、合规治理与主权算力政策」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:AI智能体、法律归责、OpenAI、Hugging Face,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始信息发布自一手权威信源「arXiv: OpenAI」(Tier-1 一手发布),具备高确定性参考价值。