这场争论表面上在问自主武器是否应该杀人，但实际上自主杀伤能力早已存在。Michael Horowitz 指出，美国政府部署自主武器系统已达 40 年；以雷达导引飞弹为例，人类在其大部分飞行流程中早被移出决策回路。最新变化是 AI 的导入，且已在乌克兰等战场出现部分自主应用，显示技术趋势正从传统自动化走向更高层级的模型驱动自主。
核心分歧不在于是否使用 AI，而在于人类应在何时、以何种强度介入致命武力。Dario Amodei 主张 AI 不应用于「完全自主」武器，也就是无人类参与即可自行选定目标并执行打击的系统。Jack Shanahan（退役三星将领）同样认为，现阶段任何 LLM 都不适合用于完全致命自主武器。其风险集中于目标识别与法规遵循：系统可能无法可靠区分坦克、战斗员与平民，或无法辨识投降，且模型输出的不可预测性会放大误判机率。
另一方强调战场通讯中断时的作战现实。Palmer Luckey 认为，若机器人不能自主决定打击目标，敌方只需干扰讯号即可瘫痪能力，并主张未来战争将由自主无人机主导。与此同时，Amodei 在 1 月文章警告，若出现「数百万至数十亿」全自动武装无人机群，可能形成几乎不可战胜的军事力量，甚至被用于国内高压监控。整体而言，争论正沿两条统计尺度升级：一是从 40 年既有自主系统延伸到前沿模型，二是从单点打击扩张到百万至十亿级蜂群规模。
This debate appears to ask whether autonomous weapons should be allowed to kill people, but in practice autonomous lethal capability has existed for a long time. Michael Horowitz notes that the US government has fielded autonomous weapons systems for 40 years; radar-guided missiles already remove humans from much of the decision loop during flight. The newer shift is AI integration, and partially autonomous use has already appeared in battlefields such as Ukraine, indicating a trend from legacy automation toward higher-level model-driven autonomy.
The central divide is not whether AI is used, but when and how strongly humans must remain in lethal-force decisions. Dario Amodei argues that AI should not be used for “fully autonomous” weapons, meaning systems that can choose targets and execute strikes without human involvement. Jack Shanahan, a retired three-star general, similarly argues that no current LLM is suitable for fully lethal autonomous weapons. The risk profile centers on targeting and legal compliance: systems may fail to reliably distinguish tanks, combatants, and civilians, or fail to recognize surrender, while model unpredictability increases the probability of catastrophic error.
The opposing view emphasizes battlefield conditions when communications fail. Palmer Luckey argues that if robots cannot autonomously decide targets, an adversary can disable capability simply by jamming signals, and he frames autonomous drones as the future of warfare. At the same time, Amodei warned in a January essay that swarms of “millions or billions” of fully automated armed drones could create an effectively unbeatable force and enable domestic repression through persistent surveillance. Overall, the argument is scaling along two quantitative axes: from 40 years of existing autonomy to frontier-model control, and from single-platform effects to swarm sizes in the million-to-billion range.