Authors: Zhang Jiadong (张家栋) and Yao Yani (姚亚妮)
Journal: International Review (《国际观察》), No. 4, pp. 88–113
Published: 2026
PDF:
生成式人工智能的崛起正在重塑国际安全边界，但其应用中的系统性责任规避问题尚未得到充分关注。本文将生成式人工智能避责定义为行为体有意识地利用技术特性规避、转移或淡化自身责任的行为模式，具体表现为责任推诿多元化、责任归因技术化与责任稀释三重态势。文章将该现象归纳为技术特殊性、行为体认知偏差与制度环境滞后三重机制叠加的结果。技术层面，算法黑箱深度动态化、输出不确定性及责任链条多极化，使因果回溯与归责难以实现；认知层面，技术中立迷思、过度信任与责任意识稀释，驱动行为体主动“去责任化”；制度层面，国际法规范建构缓慢、各国标准碎片化与开源生态责任虚化，为避责提供了规则套利空间。上述三重机制相互作用，给国际安全带来一系列挑战，包括作战决策归责失效使人道法归责路径受阻，网络攻击溯源判定失灵弱化威慑机制，内容操纵责任失锁引发认知攻击追责困境，规则共识信任失稳导致安全治理碎片化等。为有效规制这一行为，国际社会需确立基本规制原则，重构责任分配的制度架构，建立国际透明与问责机制，深化全球治理的制度协调。
关键词： 生成式人工智能；责任规避；国际安全；技术归责；责任分配
The rise of generative artificial intelligence is reshaping international security, yet systematic responsibility avoidance in its application remains insufficiently examined. This article defines such avoidance as actors’ deliberate exploitation of technological characteristics to evade, shift or downplay their responsibilities. It takes three forms: buck-passing among multiple actors, attributing responsibility to technology and diluting responsibility. The article explains this phenomenon through the interaction of technological characteristics, actors’ cognitive biases and institutional delays. Technologically, increasingly complex and dynamic algorithmic black boxes, uncertain outputs and fragmented chains of responsibility impede causal tracing and accountability. Cognitively, the myth of technological neutrality, excessive trust and a diminished sense of responsibility encourage actors to absolve themselves of responsibility. Institutionally, the slow development of international legal norms, fragmented national standards and weakened accountability within open-source ecosystems create opportunities for regulatory arbitrage. Together, these mechanisms pose several international security challenges. Failures of accountability in combat decisions obstruct responsibility attribution under humanitarian law. Difficulties tracing cyberattacks weaken deterrence. Gaps in responsibility for content manipulation impede accountability for cognitive attacks, while declining trust in shared rules fragments security governance. To address these problems, the international community should establish basic regulatory principles, redesign institutional arrangements for allocating responsibility, create international transparency and accountability mechanisms and strengthen institutional coordination in global governance.
Keywords: generative artificial intelligence; responsibility avoidance; international security; technological attribution of responsibility; allocation of responsibility