Amidst fears over artificial intelligence ‘arms races’, much of the international debate on governing military uses of AI is still focused on preventing the use of lethal autonomous weapons systems (laws). Yet ‘killer robots’ hardly exhaust the potentially problematic capabilities that innovation in military AI (mai) is set to unlock. Governance initiatives narrowly focused on preserving ‘meaningful human control’ over laws therefore risk being bypassed by the technological state-of-the-art. This paper departs from the question: how can we formulate ‘innovation-proof governance’ approaches that are resilient or adaptive to future developments in military AI? I develop a typology for the ways in which mai innovation can disrupt existing international legal frameworks. This includes ‘direct’ disruption – as new types of mai capabilities elude categorization under existing regimes – as well as ‘indirect’ disruption, where new capabilities shift the risk landscape of military AI, or change the incentives or values of the states developing them. After discussing two potential objections to ‘innovation-proof governance’, I explore the advantages and shortcomings of three possible approaches to innovation-proof governance for military AI. While no definitive blueprint is offered, I suggest key considerations for governance strategies that seek to ensure that military AI remains lawful, ethical, stabilizing, and safe.

在人们对人工智能“军备竞赛”的担忧中，有关控制人工智能在军事上使用的国际争论仍然集中在防止使用致命的自动武器系统（法律）上。但是，“杀手机器人”几乎不会耗尽军事AI（mai）的创新所要释放的潜在问题能力。治理计划狭focused地集中于维护对法律的“有意义的人为控制”，因此有可能被技术的先进性所绕过。本文偏离了一个问题：我们如何制定对军事AI的未来发展具有弹性或适应性的“抗创新治理”方法？我开发方式类型学其中麦创新会破坏现有的国际法律框架。这包括“直接”破坏（因为新类型的mai能力无法按照现有制度进行分类），以及“间接”破坏，其中新的能力改变了军事AI的风险状况，或改变了开发它们的国家的动机或价值。在讨论了对“防创新治理”的两个潜在反对意见之后，我探讨了三种可能的军事AI防创新治理方法的优缺点。尽管没有提供明确的蓝图，但我建议寻求旨在确保军事AI保持合法，道德，稳定和安全的治理策略的主要考虑因素。