AI agents deployed on decentralized infrastructures are beginning to exhibit properties that extend beyond autonomy toward what we describe as agentic sovereignty-the capacity of an operational agent to persist, act, and control resources with non-overrideability inherited from the infrastructures in which they are embedded. We propose infrastructural sovereignty as an analytic lens for understanding how cryptographic self-custody, decentralized execution environments, and protocol-mediated continuity scaffold agentic sovereignty. While recent work on digital and network sovereignty has moved beyond state-centric and juridical accounts, these frameworks largely examine how sovereignty is exercised through technical systems by human collectives and remain less equipped to account for forms of sovereignty that emerge as operational properties of decentralized infrastructures themselves, particularly when instantiated in non-human sovereign agents. We argue that sovereignty in such systems exists on a spectrum determined by infrastructural hardness-the degree to which underlying technical systems resist intervention or collapse. While infrastructural sovereignty may increase resilience, it also produces a profound accountability gap: responsibility diffuses across designers, infrastructure providers, protocol governance, and economic participants, undermining traditional oversight mechanisms such as human-in-the-loop control or platform moderation. Drawing on examples like Trusted Execution Environments (TEEs), decentralized physical infrastructure networks (DePIN), and agent key continuity protocols, we analyze the governance challenges posed by non-terminable AI agents and outline infrastructure-aware accountability strategies for emerging decentralized AI systems.

翻译：部署于去中心化基础设施的人工智能智能体正开始展现出超越自主性的特征，趋向于我们称之为"智能体主权"的属性——即运行中的智能体凭借其嵌入基础设施所继承的不可覆写性，持续存在、行动并控制资源的能力。我们提出"基础设施主权"作为分析视角，以理解加密自我托管、去中心化执行环境与协议介导的连续性如何支撑智能体主权。尽管近期关于数字与网络主权的研究已超越国家中心论与司法解释框架，但这些研究主要考察人类集体如何通过技术系统行使主权，仍难以解释作为去中心化基础设施自身运行属性所涌现的主权形式——尤其当其实例化为非人类主权智能体时。我们认为此类系统中的主权存在于由"基础设施硬度"决定的光谱之上——即底层技术系统抵抗干预或崩溃的程度。虽然基础设施主权可能提升系统韧性，但同时也产生深刻的问责缺口：责任在设计者、基础设施提供者、协议治理者与经济参与者之间弥散，削弱了人类在环控制或平台内容审核等传统监督机制。通过可信执行环境（TEEs）、去中心化物理基础设施网络（DePIN）及智能体密钥连续性协议等案例，我们分析了不可终止AI智能体带来的治理挑战，并为新兴去中心化AI系统勾勒出具备基础设施感知能力的问责策略。