U.S. Military strikes 3,000 targets in Iran with AI support, but oversight remains "underinvested"

Human analysts can examine at most 4 percent of the intelligence material that is typically collected, according to U.S. officers who have worked in the field. "The biggest immediate impact of AI is in intelligence," Israeli Col. Yishai Kohn, the defense ministry's head of planning, economics and IT, told the WSJ. "Many potential missions simply never happened because the manpower didn't exist" to assess vital intelligence, Kohn said.

AI-powered machine vision can now quickly find vast numbers of targets, with the ability to single out specific models of aircraft or vehicles, the report says. Users can drill into results with queries, such as identifying every missile launcher located near a hospital, or set alerts: "Tell me every time someone takes a photo near this military base."

The Pentagon as a "hardware company in a software era"

The Pentagon's first AI chief, retired Air Force Lt. Gen. Jack Shanahan, described a fundamental problem to the WSJ: "The Department of Defense was built as a hardware company in the industrial age, and it has struggled to become a digital company in a software-centric era." Shanahan had overseen the AI-powered Project Maven in Iraq almost a decade ago.

The Pentagon's AI tools are similar to ChatGPT and other mass-market large language models, but limited to warfare and trained to tackle specific tasks using relevant information, the WSJ reports. Yet war remains "among the most chaotic and complex human endeavors," posing unique problems for even the cutting edge of robotic thinking. One reason: much of the available training data is out of date or unclear.

Out of every 10 people in the military, at most two face combat, according to the report. Up to 90 percent of personnel are in support roles. Precisely these areas, intelligence, mission planning, and logistics, are ripe for AI-driven efficiency gains.

20 people replace 2,000

The WSJ confirms a figure previously known from a Georgetown study: The U.S. Army's 18th Airborne Corps, using software from Palantir Technologies in a continuing string of exercises dubbed Scarlet Dragon, matched its own record from Iraq as the military's most efficient targeting operation ever. According to Emelia Probasco, a senior fellow at Georgetown University's Center for Security and Emerging Technology, the corps achieved that with only 20 people, compared with more than 2,000 staffers employed in Iraq for the same task.

NATO is also already using AI operationally. French Adm. Pierre Vandier, NATO's top officer for digital transformation, told the WSJ that alliance members are using AI to track Russia's shadow fleet of tankers, scanning millions of square miles several times a day for vessels that are illegally transferring fuel at sea. "The number of targets you can nominate through AI is just skyrocketing," Vandier said.

From weeks of planning to days

Mission planning that previously took weeks and filled "paper-stuffed binders" can potentially be done in days with AI, military leaders say. Planning any military assault typically brings together around 40 subject-matter specialists, including intelligence officers, combat commanders, weapons experts, and logistics managers. Every change has cascading effects: if intelligence shifts a bombing target to a more distant objective, commanders may opt to use different aircraft or weapons, which in turn can affect crew rostering, flight planning, and fuel consumption. AI can now process these complex interactions instantaneously, according to the WSJ.

For post-strike battle damage assessment, the military relies on so-called sensor fusion: AI simultaneously digests visuals, radar, heat signatures, and mass spectroscopy to synthesize a list of possible conclusions. Fast analysis of where attacks succeeded or failed in turn helps refine lists of subsequent targets.

Warning against overtrusting the machine

Despite the efficiency gains, several military officials cited in the WSJ report warn that the technology's capabilities risk prompting an overreliance on AI-generated information. This trend is internally described with the phrase "The computer said to do this."

Emelia Probasco, who held various posts in the Navy before becoming a senior fellow at Georgetown University, called the danger of offloading decisions to AI "a serious concern." As with other weapons systems, safeguards must be implemented to limit risks, she told the WSJ. "That infrastructure is underinvested in now," Probasco said. She was referring to the oversight and control mechanisms meant to ensure that humans critically review AI recommendations before they lead to lethal decisions.

The WSJ also reports that U.S. military investigators believe American forces likely were responsible for a strike on the war's first day that killed dozens of children at a girls' elementary school in Iran. This language is more definitive than earlier reports, which described an ongoing review.