Congress Confronts Reality of AI War Losses

A central promise of military artificial intelligence is that it will make warfare more precise and thereby protect civilians. This argument, long championed by tech executives and defense officials, suggests that algorithms can distinguish combatants from non-combatants with a level of accuracy human operators cannot match. However, a recent congressional hearing has put this optimistic narrative under intense scrutiny, with witnesses arguing that the data on the ground tells a different story.

During a session hosted by the Tom Lantos Human Rights Commission, Rep. Jim McGovern noted that the government has officially argued for AI’s role in reducing collateral damage. Yet, experts testifying before Congress pointed to recent conflicts in Iran, Ukraine, and Gaza as evidence that this reduction has not materialized. Instead of fewer casualties, the hearing highlighted a surge in the number of targets struck, often with devastating consequences for civilian populations, challenging the core ethical justification for these systems.

Accuracy claims clash with field evidence

Proponents of AI targeting systems often cite accuracy rates as high as 90%. Steven Feldstein, a senior fellow at the Carnegie Endowment for International Peace, pushed back against this figure, suggesting that real-world performance may be closer to 50%. This discrepancy is critical because it undermines the safety net that companies like Palantir claim their software provides. As reported by GN technics/ai (en-US), the gap between marketing claims and operational reality remains a significant point of contention among defense analysts.

Deborah Brown from Human Rights Watch emphasized that even if targeting is precise, the overall scale of operations can lead to massive civilian harm. She pointed to the high number of deaths in Gaza and Iran as evidence that AI is not limiting harm. The argument is that speed and scale, rather than precision, are driving the outcomes of these conflicts, resulting in a net increase in civilian casualties despite the technological sophistication of the weapons used.

Scale of strikes outpaces precision

The U.S. military struck 13,000 targets in the first 38 days of recent operations, aided by AI systems like Palantir’s Maven Smart System. Feldstein explained that even with improved accuracy, the sheer volume of targets means the total number of civilians harmed is likely higher. This is a key trade-off: while individual shots may be more accurate, the aggregate effect of automated, high-volume targeting can lead to greater collateral damage than traditional methods.

Amanda Klasing of Amnesty International USA added that accuracy is irrelevant if the principle of proportionality is ignored. She cited instances where massive bombs were dropped on residential buildings, arguing that no level of targeting precision justifies such disproportionate force. The technology executes the rules of engagement set by humans, which may not prioritize civilian protection, effectively automating decisions that carry significant ethical weight.

Human oversight is being reduced

As AI systems take on more targeting roles, the human structures designed to mitigate harm are shrinking. The Pentagon’s Civilian Protection Center of Excellence, established to train military commands on reducing civilian harm, is cutting its staff from 40 to just nine by the end of the year. This reduction in oversight capacity raises concerns that the systems are being deployed without the necessary human safeguards to ensure compliance with international humanitarian law.

Critics argue that this hollowing out of training and oversight contradicts the stated goal of more humane warfare. If the systems are programmed to accept a certain threshold of civilian deaths, the technology becomes a tool for scaling up conflict rather than refining it. The debate in Congress is now shifting from whether AI can reduce harm to whether the current military framework allows for any meaningful reduction at all.