Pentagon’s Push for AI on the Battlefield Draws Warnings From Top Commanders

The U.S. military is racing to put artificial intelligence in charge of finding and striking targets and some of its own top commanders are telling it to slow down. Admiral Frank Bradley, who leads U.S. Special Operations Command, warned at a recent conference in Tampa, Florida, that troops “have to be very careful about how we come to (AI’s) employment and its inspiration into the delivery of lethality.” The man saying this oversees the military’s most dangerous and secretive operations.

Bradley chairs the same command whose boss, Defense Secretary Pete Hegseth, has made AI expansion one of his defining priorities. Hegseth told SpaceX employees in January that he would reject any AI models “that won’t allow you to fight wars,” calling for systems that operate “without ideological constraints that limit lawful military applications.” The vision Hegseth is pursuing and the warning Bradley is issuing are coming from the same chain of command.

Bradley has not ruled out a future where AI selects military targets. His concern is about precision. He said humans must have confidence that the technology “is going to deliver violence only where we intend it to be delivered.” The gap between that standard and the Pentagon’s current pace of adoption is where the real tension lives and it goes well beyond one admiral’s remarks at a Florida conference.

Inside the Military, AI Means Two Very Different Things

When Pentagon officials were asked to respond to Bradley’s concerns, an anonymous official said the focus is on building “functional battlefield tools” that help troops identify and process targets faster, directly speeding up strikes. That is a markedly different picture from what Bradley’s own command is describing. Officials at Special Operations Command frame AI as a way to give operators more time, not more firepower.

Sgt. Maj. Andrew Krogman, the top enlisted officer at Special Operations Command, described AI as handling administrative work and modernizing how the command operates. Melissa Johnson, the command’s top acquisition official, said the technology should be “reducing the cognitive workload on mundane tasks.” She added that the goal is to enhance operator judgment, not replace it. Two officials from the same command, offering two very different definitions of what the technology is actually for.

Helen Toner, interim executive director at Georgetown University’s Center for Security and Emerging Technology, said both descriptions are accurate. The military does use AI for bureaucratic efficiency, and it also uses AI to find and strike targets. Her center published a case study showing the Army’s 18th Airborne Corps achieved the targeting accuracy of the best unit in recent American history using 2,000 fewer service members, with AI driving the difference.

AI Is Already on the Battlefield, Whether the Debate Is Settled or Not

Lt. Gen. Michael Conley, head of Air Force Special Operations Command, told Congress in May that his troops used AI tools during the Iran war to convert top-secret intelligence to a lower classification level in seconds, making it shareable with drone operators on the ground. That is not a future capability. It already happened. The debate over guardrails is running behind the technology it is supposed to govern.

Toner, a former board member at OpenAI who was removed following a dispute with CEO Sam Altman, offered a counterpoint to public fears about military AI. According to Toner, the general public tends to underestimate how carefully the U.S. military evaluates new technology. She noted that commanders need operations to succeed, which means avoiding friendly fire, civilian casualties, and incorrect target identification just as much as it means delivering lethal force effectively.

Still, the speed of the Pentagon’s AI integration is outpacing the policy frameworks meant to control it. President Trump pulled back from signing a new AI executive order hours before the scheduled signing ceremony, citing concern that safety measures could blunt America’s competitive edge. “We’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead,” Trump told reporters. The guardrails question has become a geopolitical calculation.

A $200 Million Rupture That Exposed the Limits of AI Governance

The most direct confrontation over military AI did not happen on a battlefield. It happened in a contract dispute. After Anthropic CEO Dario Amodei refused to remove restrictions barring his company’s Claude AI from being used in fully autonomous weapons or domestic mass surveillance, the Pentagon terminated its $200 million defense contract with the company and formally designated Anthropic a supply chain risk, blocking other government contractors from working with it.

Anthropic sued, arguing the designation was illegal retaliation for its expressed views on AI ethics, not a genuine security threat. The Pentagon pivoted to Google, OpenAI, and SpaceX to fill the gap, seeking AI that could “augment warfighter decision-making in complex operational environments.” The episode drew a federal judge’s scrutiny and exposed a central unresolved question: who sets the ethical limits on AI when the government itself is the one pushing to remove them.

What the Anthropic dispute made plain is that the argument over AI safeguards is no longer abstract. It has a dollar amount, a set of named parties, and a court filing. Bradley’s warning about delivering violence “only where we intend it” and Hegseth’s demand for AI “without ideological constraints” are not two positions waiting to be reconciled. They are two definitions of accountability that the U.S. military has not yet decided between — and the answer will shape how American power is exercised for decades.