US Military Defies Trump Ban to Use Anthropic’s Claude in Iran Strikes

The intersection of artificial intelligence and kinetic warfare has reached a critical flashpoint as the US military reportedly deployed Anthropic’s Claude AI during recent strikes on Iranian targets, directly contravening an executive order from President Donald Trump. This defiance marks an unprecedented rift between the Commander-in-Chief and the Pentagon’s operational leadership, exposing the technical and political complexities of 'unplugging' deeply integrated software during active combat operations. The controversy began when President Trump signed an order hours before the strikes, labeling Anthropic a 'national security risk' and a 'radical Left' entity, largely due to the company’s refusal to grant the military unrestricted rights to its technology or allow its use for fully autonomous lethal decisions.

At the heart of the dispute is Anthropic’s 'Constitutional AI' framework, which embeds specific ethical principles into the model's core logic. During negotiations for a major Pentagon contract, Anthropic leadership reportedly insisted on safeguards against mass surveillance and autonomous killing, a stance that Defense Secretary Pete Hegseth and other administration officials viewed as a supply chain risk comparable to hostile foreign firms like Huawei. Despite this high-level political designation, US Central Command (CENTCOM) continued to utilize Claude for intelligence analysis, target identification, and real-time battlefield simulations during the Iran operation. Military officials argued that the AI was so deeply embedded into their existing intelligence platforms—likely including systems integrated by Palantir—that there was no immediate substitute available to maintain operational tempo.

“

Despite this high-level political designation, US Central Command (CENTCOM) continued to utilize Claude for intelligence analysis, target identification, and real-time battlefield simulations during the Iran operation.

This incident underscores the 'sticky' nature of AI infrastructure in defense. Once a large language model (LLM) is integrated into a military workflow, it becomes more than just a tool; it becomes the cognitive layer through which commanders interpret complex data. The Pentagon’s inability to comply with the ban on such short notice suggests that the transition to alternative models will be a multi-month, if not multi-year, technical challenge. This reality complicates the administration's efforts to pivot toward more 'unconstrained' AI providers who are willing to operate without the ethical guardrails championed by Anthropic’s founders.