By: Evan Teates
The United States and China are locked in an artificial intelligence arms race, competing to outpace one another in a technology that is increasingly shaping modern warfare. As it was in the nuclear arms race of the Cold War, neither side is incentivized to slow its AI development, as doing so risks falling behind on the technological front.
As countries increasingly integrate AI into military strategy and operations, the ethical dilemma is how far AI should be allowed to influence military decision-making before meaningful human control is lost.
This risk stems from militaries' pursuit of greater speed. AI can process and generate information at rates beyond human capabilities. However, as military systems become faster and more autonomous, the window for humans to correct an erroneous decision becomes increasingly narrow.
The supporting argument for military AI, as emphasized by U.S. defense officials, centers on improvements in speed and target discrimination, as restricting its use could mean surrendering strategic advantage. However, this does not eliminate error and the need for human judgment. It raises a question of whether the pursuit of AI military means comes at the expense of human oversight.
The United States has since made AI acceleration a military priority. In January 2026, the Pentagon released its AI strategy, directing an “AI-first” warfighting force to achieve “Military AI Dominance.” Likewise, China has prioritized AI as part of its military modernization, with President Xi Jinping ordering the People’s Liberation Army to accelerate the deployment of “unmanned intelligent technologies” to gain a strategic advantage.
This shift is already visible in the Russo-Ukrainian war, where both sides are advancing AI drone capabilities that assist with target recognition and strikes, with the potential to accelerate the military “kill chain,” or the process of locating a target, relaying the data, and carrying out a strike. Ukraine’s Ministry of Defense identifies the acceleration of this process as a key incentive for AI integration.
However, accelerating the kill chain narrows the window for human oversight, as decision-making becomes increasingly compressed with limited time to verify the assessments of an AI military system. Experts warn that these systems can consist of multiple algorithms chained together, each with its own flaws that can feed into subsequent stages of the kill chain, thereby increasing the risk for civilian casualties. The UN Human Rights Monitoring Mission in Ukraine has already documented the devastating civilian toll of drone attacks, and as AI becomes more integrated into drone warfare, that risk becomes amplified.
Yet, technical failures are not the only risk. Human operators can also place too much trust in an AI system's recommendations, even when those recommendations are prone to errors. In the Gaza war, AI has assisted Israeli targeting through a system known as “Lavender.”
A 2024 investigation by +972 Magazine and Local Call reported that Lavender identified approximately 37,000 Palestinians as suspected militants for potential strikes. According to the investigation, operators sometimes spent merely 20 seconds reviewing a target before authorizing a strike, despite a reported error rate of 10%. Military personnel reportedly treated its outputs “as if it were a human decision” even with the prevalence of target misidentification. This illustrates automation bias in human oversight, which influences targeting decisions that lead to civilian deaths.
Furthermore, military AI poses a risk of rapid escalation. A “flash-war” scenario could emerge when AI systems trigger escalation faster than humans can respond. The Stockholm International Peace Research Institute warns that AI can increase the risk of nuclear escalation even when integrated into conventional military systems. As in the previous examples, these systems compress decision-making, leaving less room for human judgment. In a nuclear crisis, that loss of time could have catastrophic consequences.
Reducing these risks does not require the United States and China to abandon strategic competition. The Cold War demonstrated that strategic competition need not preclude arms control. For instance, the United States and Soviet Union established a direct communications hotline in 1963 following the Cuban Missile Crisis to reduce the risk of miscalculation during future nuclear crises.
More recently, in November 2024, President Biden and President Xi affirmed that humans, not AI, should retain a strict control over nuclear weapons. This understanding provides a starting point for cooperation. In September 2026, US and Chinese security experts proposed nuclear-style safeguards for military AI, including provisions for human control over consequential military actions and the establishment of communication channels for AI-related incidents that mirror the 1963 nuclear arms hotline. These proposals have not yet been formally adopted.
Moreover, the United States is well positioned to help advance these efforts. It controls an estimated 74% of global high-end AI computing capacity, giving it significant influence over AI development. The United States should use this position to lead negotiations with China and other nations to implement the safeguards proposed by US and Chinese security experts in an international framework, thus ensuring that the pursuit of military speed and strategic advantage does not come at the expense of meaningful human control.
While the United States and China will continue to propel the race for military AI superiority with a continued effort to develop that technology, meaningful human control becomes an ever more essential component of regulating the race. The goal should not be to slow AI development for its own sake, but ensuring that the pursuit of military advantage never outpaces human judgment.