The United States military used Anthropic’s Claude model during its large-scale airstrikes on Iran, according to a report in The Wall Street Journal that was picked up by several outlets on February 28 and March 1. The reported use came only hours after President Donald Trump directed federal agencies to stop working with the company.
The episode captures a contradiction at the center of the current debate over artificial intelligence in defense: the US government is trying to distance itself from one of the leading AI developers at the same time that its forces are depending on that developer’s technology in live operations.
What was reported
According to the Journal’s reporting, US Central Command, which oversees military operations in the Middle East, has been using Claude for intelligence analysis, target selection and simulations of combat scenarios. Other reports have said the same model was used in the January operation in which US forces detained Venezuela’s former president, Nicolas Maduro, although that detail has not been independently confirmed.
Taken together, the reports suggest that AI tools are no longer confined to back-office analysis. They are being applied across what military planners call the “kill chain,” the sequence of steps from identifying a target to deciding whether to strike it.
A dispute over acceptable use
The reported battlefield use follows weeks of friction between Anthropic and the Department of Defense. The dispute centered on whether Anthropic would allow its models to be used for “all lawful purposes” by the military, including fully autonomous weapons and mass domestic surveillance. Anthropic refused to drop those restrictions, and negotiations with defense officials broke down.
On February 27, Trump directed federal agencies to end their use of Anthropic’s products within six months, and the Pentagon designated the company a supply chain risk, a label normally applied to foreign adversaries’ suppliers. Anthropic has said it intends to challenge the designation in court. The company has argued that its limits are narrow and that it continues to support lawful national security work.
At the same time, the Defense Department has reached agreements with other AI providers, including OpenAI and xAI, to bring their models into military systems. That signals that the government’s quarrel is with one supplier’s terms, not with the use of AI in warfare itself.
Why it matters
The sequence of events raises practical and ethical questions that defense planners, lawmakers and technology companies will have to answer. The practical one is dependence: if a frontier model is woven into targeting and analysis workflows, switching it off on short notice is harder than issuing an order. The reported use of Claude during the Iran strikes, after the phase-out directive, illustrates that gap.
The ethical and legal questions are older but more urgent. When AI systems inform decisions about who or what to strike, the standards for human oversight, accountability for errors and the handling of civilian harm need to be explicit. Anthropic’s stance, and the government’s reaction to it, show that there is no settled consensus on where those lines should be drawn, or on who gets to draw them.
What is clear is that AI has moved from the periphery of military planning to its core. The companies that build these systems and the governments that deploy them are now negotiating the terms of that relationship in public, and the outcome will shape how the next generation of conflicts is fought.
