AI Oversight Erosion: From Military Targeting to Data Centers

A Charlotte Observer opinion piece published in early 2026 draws a direct line between AI-assisted military targeting and the erosion of institutional safeguards that once prevented catastrophic errors. The author, Patricia Setari, a U.S. Army veteran with 30 years of national security experience, argues that speed without accountability transforms mistakes into irreparable harm.

Setari's argument rests on two specific incidents separated by 27 years. In 1999, American bombs struck the Chinese embassy in Belgrade, killing three people. The targeting data was outdated, but the bombs hit exactly where they were aimed. The failure was intelligence, not precision. In February 2026, American forces struck the Shajareh Tayyebeh primary school in Minab, Iran, killing at least 156 people, most of them schoolchildren between 7 and 12 years old. The building had been misclassified in a Defense Intelligence Agency database that hadn't been updated to reflect its separation from an adjacent military compound.

The difference between these two events is the role of artificial intelligence. According to Setari's account in the Charlotte Observer opinion piece, an AI-assisted targeting system called Maven generated approximately 1,000 targets in the first 24 hours of Operation Epic Fury with a staff of roughly 20 people. In 1999, building a target list required thousands of analysts working over days and weeks, with lawyers at every step reviewing compliance with the laws of war.

The oversight infrastructure has been deliberately dismantled. Pete Hegseth, U.S. Secretary of Defense, publicly declared there would be "no stupid rules of engagement." The Pentagon's top military lawyers had been fired months earlier. Their entire job was to ask the hard questions before bombs fall. The president had stated openly that international law didn't apply to him.

What Setari learned from Belgrade is that safeguards aren't bureaucratic red tape. They are the difference between a mistake you can learn from and a war crime you can't take back. They exist precisely because wars move fast, databases go stale and human beings make errors. Slow the targeting cycle down, and you catch things. Speed it up without accountability, and you get Minab.

The connection between military AI and civilian infrastructure is more direct than most observers acknowledge. The energy demands driving military AI are the same demands bringing hyperscale data centers into cities around the U.S., often without public input or accountability. A proposed "hyperscale" data center in Box Elder County, Utah, dubbed the Stratos Project and backed by celebrity venture capitalist Kevin O'Leary, will devour up to nine gigawatts of energy, more than double the electricity used by the entire state.

Robert Davies, a physics professor at Utah State University, calculated that the project would be the equivalent of about 23 atom bombs worth of energy dumped into the local environment every single day. The facility will produce another 7 to 8 gigawatts of energy in the form of waste heat. That brings its entire thermal load to a jaw-dropping 16 gigawatts. The valley, Davies predicts, will become yet another desiccated landscape that adds to the area's dust problem as the shrinking of the Great Salt Lake exposes more lakebed.

This environmental impact report from Futurism illustrates the physical reality of AI infrastructure. The facility will be equal to about 2,000 Walmart Supercenters in acreage, but its energy footprint will be that of 40,000 Walmart supercenters, or 2,000 Walmarts stacked 20 deep. The heat island effect could spike local temperatures by five degrees Fahrenheit during the day, and 28 degrees at night.

The military's AI spending provides additional context. The Defense Department has allocated at least some $75 billion to AI-driven programs since 2016, as explained in a Brennan Center report. The actual total could be far larger, as this number does not cover programs that are secret or those where the extent of AI use is unclear. The military is investing heavily in the development of autonomous weapons, which can select targets and take lethal action with varying degrees of human involvement. The Pentagon requested $13.4 billion for these types of systems for 2026 alone.

According to the Brennan Center analysis available at their research portal, much of the Pentagon's AI-related procurement dollars have gone to data analytics giant Palantir and Anduril, which manufactures AI-powered drones. Palantir and Anduril recorded their largest-ever annual defense revenue in 2025 — $903 million and $912 million, respectively. Palantir is the lead contractor on the Maven Smart System, which has been used in Iran, Iraq, Syria, Ukraine, and Yemen.

The rush to adopt AI threatens to displace human expertise and judgment in life-and-death decisions, jeopardizing troops and civilians alike. Anyone who has used AI chatbots knows that they frequently make mistakes, both obvious errors and ones that are harder to detect. AI is prone to inaccuracy in the military context as well. In 2024, Maven's algorithms were reportedly able to correctly identify a tank in good weather about 60 percent of the time, and its accuracy dropped to only 30 percent in snowy conditions.

Foundation models also generate false or misleading analysis while making it sound persuasive. This makes it more likely that commanders and analysts will accept their recommendations, especially during the heat of war. Even if humans are making final decisions, relying on AI for target selection or justification can lead to incorrect outcomes — and in military situations, these mistakes can have deadly consequences.

The dispute between the Pentagon and Anthropic raises additional questions about whether the military's deployment of AI technology is effective, safe, and lawful. Anthropic wanted the military to promise that it would not use its AI model, Claude, in weapons that can identify and fire on targets without human input — commonly referred to as "fully autonomous weapons." The Pentagon refused to agree to these restrictions, then blacklisted Anthropic from defense contracting by designating the company a "supply chain risk."

What's happening with military AI mirrors what's happening with civilian infrastructure. Governments are now classifying massive AI data centers as "military operations," quietly stripping communities of any power to stop them. Local control is disappearing fast. And it's being replaced by national security justifications as residents are locked out of decisions that are quickly reshaping entire communities.

The physical reality of this technology is worth noting. When you interact with AI systems, you're clicking through interfaces that process data at speeds human brains can't comprehend. The friction of human review — the time it takes to read a document, verify a source, ask a question — gets compressed into milliseconds. That compression feels efficient until it doesn't (and when it fails, the consequences are measured in human lives).

Setari's conclusion is blunt: we cannot move at machine speed toward a future we haven't thought through. She's seen what happens when we do. The safeguards exist for a reason. They're not obstacles to progress. They're the difference between a system that works and one that kills children in schools.

Whether users actually pay for AI efficiency with their safety remains the real question. The technology works. The oversight doesn't. And when those two things collide, the people who pay the price aren't the ones who built the system.

The children in Minab deserved the protection our safeguards exist to provide. So do the men and women still serving. So do we. Whether Congress actually uses its constitutional authority over war powers is another matter entirely.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn