The Pentagon is racing to integrate AI into virtually every aspect of the U.S. military. While AI has the potential to strengthen national security, the Trump Administration’s rush to deploy it on the battlefield threatens to drive up long-term taxpayer costs and could cause catastrophic mistakes.

The Pentagon’s summary of its FY2027 budget request describes a “historic” $58.5 billion in taxpayer funding for AI. The total cost of military AI is likely far higher than that figure, given that programs across the department are working to integrate AI regardless of whether it’s a primary focus. For example, in addition to the $58.5 billion, the FY2027 budget request includes $2.7 billion for AI-powered Collaborative Combat Aircraft, and $54.6 billion for the Defense Autonomous Working Group, an AI-driven drone and autonomous weapons program.

Defense Secretary Pete Hegseth has repeatedly emphasized his “unrelenting focus on speed” when it comes to AI. When he launched the Department of Defense’s AI Acceleration Strategy in January, he declared a “wartime approach” to anything that could slow AI acceleration, calling them “operational risks” that they will be “blowing up.”

A Government Accountability Office (GAO) report released in October suggests that Secretary Hegseth’s AI approach is leaving major oversight gaps. GAO found that 29 of the 64 efforts laid out in the department’s own plan for responsible AI implementation remain incomplete. Moreover, just three of the ten authorized positions overseeing responsible AI implementation were filled as of February 2026.

The taxpayer costs of military AI are also ballooning. A separate GAO report from April found that the Pentagon often underestimates the lifetime costs of military AI by failing to account for costs involved in constantly retraining AI models, adding additional computing power as demands on AI models grow, and paying for continuous access to cloud-based infrastructure.

With billions of taxpayer dollars and national security at stake, the Pentagon must ensure its new AI systems perform reliably. Several recent incidents suggest the Department of Defense has not done enough to ensure taxpayers’ massive investments in military AI are creating safe and effective systems that actually improve national security outcomes.

Earlier this year, an AI-assisted intelligence report falsely claimed a Chinese ship in the Middle East was carrying components of a nuclear weapons program. The U.S. military prepared to intercept the Chinese vessel, with personnel readying to board and aircraft already airborne. Fortunately, someone at the Pentagon drilled down on the findings of the AI report and concluded the information was wrong before the interception happened. One source told CNN that the report was “entirely false” and “almost started a war.”

Other consequences of the administration’s approach to military AI have gone beyond close calls, with tragic results. Bloomberg found that an overreliance on AI led to the February 28 missile strike on an elementary school in Minab, Iran. The attack killed more than 150 people, including at least 123 children. According to Bloomberg, the school was recommended as a target by an AI system that likely relied on outdated intelligence in recommending the target. The U.S. Central Command’s civilian-harm mitigation team, which the administration had reduced from ten people to one, did not review the target before the strike.

The Minab bombing gets to the heart of the taxpayer issues with Secretary Hegseth’s AI approach. Taxpayers have shelled out billions for military AI, and yet the systems and the personnel using them failed to prevent a catastrophic mistake. Now the administration is requesting billions more taxpayer dollars to continue integrating unreliable, unproven systems.

In his January speech on AI, Secretary Hegseth said “Our department doesn’t accept failure in the past, and so we never fail, which means we never learn. We’re flipping that dynamic.” Taxpayers can arguably tolerate some risk of failure at the research and development stage in pursuit of new technologies. But deploying unproven systems on the battlefield is a different kind of risk. When military AI fails during combat operations, taxpayer dollars aren’t the only casualty—servicemembers, civilians, and national security are all at risk.

Rather than learning from its mistakes and working to implement stronger safeguards to prevent similar outcomes in the future, the Pentagon is doubling down on its “move-fast and break things” approach. In announcing plans to stand up a new combatant command called Autonomous Warfare Command, or AUTOWARCOM, Secretary Hegseth explained that “realistic training exercises, and when possible, the crucible of combat, will be used to compress innovation timelines and scale what works.” In other words, the plan includes deploying AI in combat before answering the most basic question that should be a guiding star in every Pentagon acquisition decision; “Does it work?”

Before spending billions more taxpayer dollars on military AI, Congress should ensure these systems undergo strong independent testing and include comprehensive lifetime cost estimates. Lawmakers should also ensure GAO’s outstanding recommendations on military AI are implemented. Lastly, Congress should implement safeguards to prevent the use of unproven AI systems in combat, to ensure that military targets identified by AI are verified by human counterparts, and to strengthen oversight of systems once they are deployed. Until those steps are taken, the Pentagon’s AI approach is a recipe for wasteful taxpayer spending and costly mistakes.