Pentagon used Anthropic’s Claude in Iran attack hours after Trump’s phase-out order

The Pentagon is still using Anthropic’s Claude, even after Trump ordered a phase-out

• US military used Anthropic's AI in major airstrike on Iran
• Pentagon ordered agencies to stop using Anthropic's AI tools
• Replacing Anthropic's Claude AI could take months

Did our AI summary help?

The U.S. military used artificial-intelligence software from San Francisco-based startup Anthropic in a major air attack against Iran hours after President Donald Trump directed federal agencies to end use of the company’s AI systems, according to people familiar with the matter who spoke to the Wall Street Journal.

Those people told the Journal that commands around the world, including U.S. Central Command in the Middle East, employ Anthropic’s Claude AI tool for tasks such as intelligence assessments, identifying targets and simulating battlefield scenarios. Centcom declined to comment on specific systems in its ongoing operations against Iran.

The use of Claude in such high-stakes missions highlights that the model is already integrated into U.S. military operations even as relations between Anthropic and the Pentagon have deteriorated sharply, the Journal’s sources said.

The standoff stems from months of disagreement between Anthropic and the Defense Department over terms for using the company’s AI models. Last week, the Trump administration ordered all federal agencies to stop working with Anthropic. It directed the Pentagon to designate the company a security threat and risk to its defense supply chain. That directive followed contract negotiations in which Anthropic refused to grant the Pentagon the right to use Claude in all lawful scenarios it might require, according to the Journal.

The conflict has prompted the Defense Department to secure alternative contracts for AI tools from other developers. The Pentagon has reached deals with the makers of OpenAI’s ChatGPT and Elon Musk’s xAI models for classified settings, but military officials and AI experts say fully replacing Claude across all systems could take months.

Anthropic was among a small group of major AI labs, alongside OpenAI, Google and xAI, that secured multiyear contracts, each worth up to around $200 million, with the Pentagon to supply advanced AI capabilities. Anthropic’s Claude model became the only one approved for use in classified military and intelligence workflows through a partnership involving Palantir and Amazon Web Services.

The model’s integration into Pentagon systems, including intelligence workflows and operations such as the January raid in Venezuela that captured President Nicolás Maduro, drew attention earlier this year and underscored its operational value within high-security contexts.

Tensions escalated in recent weeks after Defense Secretary Pete Hegseth gave Anthropic a deadline to allow unrestricted use of its AI tools for any “lawful” military purpose. Anthropic’s CEO, Dario Amodei, publicly rejected demands to drop safeguards against specific uses, framing them as ethical red lines that the company would not cross, even at the cost of government contracts.