The Pentagon says it has stopped using Anthropic's artificial intelligence tools, ending a months-long standoff over the company's refusal to strip safety guardrails from its Claude models. The confirmation, given to the BBC on October 5, comes well after the department's own phase-out deadline — and after sources said Claude was still quietly powering American military work as recently as last week.
 |
| An aerial view of the Pentagon, where officials say Anthropic's AI tools have now been phased out. |
A Statement Months After the Deadline
"The Pentagon 'has ceased the use of Anthropic products,'" a department official said in a statement to the BBC on Monday, October 5. The statement did not specify when the use actually ended — a notable omission, because the Pentagon's own timeline said it should have ended by late August.
Defense Secretary Pete Hegseth designated Anthropic a national security "supply chain risk" in February and set a six-month deadline, announced on February 27, to purge the company's tools from military systems. That put the cutoff in late August. It was not met.
Multiple people familiar with the matter told the BBC that as recently as last week, Anthropic's Claude model was still being used by the department in research, analysis, and intelligence gathering — as well as in military operations against Iran. The sources included former US defense officials and contractors who have worked closely with the Pentagon specifically on AI. They described Claude as a critical tool for sorting through vast quantities of information. It is unclear why the department took longer than its own deadline to cut ties; the details of the department cutting ties entirely had not been previously reported.
An Anthropic spokesman declined to comment when asked about the Pentagon's statement.
The Guardrail Standoff
The fight started in February, when Hegseth demanded that Anthropic remove restrictions on Claude so the military could use it for "all lawful purposes." Anthropic refused to drop its own guardrails against two things: fully autonomous lethal weapons and mass domestic surveillance of Americans.
Hegseth reportedly called Claude's capabilities "exquisite" and the company's objections "understandable" — and then blacklisted the company anyway. In March, the Pentagon formally designated Anthropic a national security supply chain risk, a label that cancelled its military contracts and barred other defense contractors from touching its models.
The designation was an unusual use of the authority against an American company. As Reuters has noted, the supply-chain-risk power had previously been applied to companies considered national security risks linked to foreign adversaries. Anthropic called the move "unprecedented and unlawful" and has sued the Trump administration to overturn it.
The company says the fallout has cost it billions of dollars in lost business — landing just as it heads toward a potential initial public offering that reporting has suggested could value it near $2 trillion.
Two Courts, Two Verdicts
The courts have split on whether the blacklisting was lawful. In August, US District Judge Rita Lin ruled the Pentagon had illegally retaliated against Anthropic for speaking out, violated its due process rights, and branded it a supply chain risk without lawful basis. That ruling remains in effect.
But on September 25, a panel of the US Court of Appeals for the District of Columbia Circuit reached the opposite conclusion in a 2-1 decision, upholding the Pentagon's designation. The court said the department had "ample support" for its conclusion that Anthropic's technology could create risks for military operations. The judges said Anthropic could alter how its Claude model responded to users, potentially restricting its use in ways the Pentagon could not control — and that the company determines the behavior of each new version, forcing the military to keep up.
"In our Republic, it is the President and the Secretary of War who must determine how best to balance" the risks, Judge Gregory Katsas wrote for the majority.
The appeals court also surfaced a striking detail: an Anthropic executive had questioned the use of Claude by the Pentagon contractor Palantir during the military operation that captured Venezuelan President Nicolas Maduro on January 3. According to Under Secretary of Defense Emil Michael, as quoted in the opinion, that objection "led to alarm" and "raised material doubts as to whether they would cause their software to stop working or cause some other disastrous action that would put our warfighters' lives in danger." Anthropic insists it cannot modify its models once delivered to the military.
 |
| Defense Secretary Pete Hegseth blacklisted Anthropic as a 'supply chain risk' in February and ordered a six-month phase-out of its AI tools. |
The Missed Deadline
The months-long lag between the late-August deadline and the October 5 confirmation is telling. As AI analyst Lauren Kahn noted, the fact that it took the Pentagon this long to remove all use of Claude across the system shows "these things are not just plug and play." Claude was deeply embedded in intelligence workflows; ripping it out — and replacing it — is a nontrivial operation.
What This Means
The Pentagon-Anthropic rupture is the most dramatic test yet of a question Washington has been circling for years: who controls the weapons-grade AI that Silicon Valley builds — the companies, or the government that buys it?
Anthropic was not some experimental vendor. It had been used by the US government and military since 2024 and was the first advanced AI company to have its tools deployed in agencies doing classified work. Its refusal to allow use for autonomous killing or mass surveillance of US citizens was core to its corporate identity — and became, in the Pentagon's view, exactly the problem. From Hegseth's standpoint, an AI provider that might one day decide its software shouldn't work for an operation is a dependency the military cannot tolerate. From Anthropic's standpoint, surrendering that veto means handing the world's most powerful military an AI with no red lines.
 |
| Anthropic's Claude model was the first advanced AI tool deployed in US agencies doing classified work, starting in 2024. |
What Happens Next
Legally, the fight is far from over. Anthropic has said it "respectfully disagree[s]" with the appeals court's ruling and is considering asking the full appeals court to review the decision. With Judge Lin's opposite ruling still on the books, the dispute could keep moving upward — and its outcome will shape how every other AI lab negotiates with the Pentagon.
Operationally, the department is pivoting. Reporting indicates the Pentagon has accelerated its reliance on alternative frontier models, with new defense contracts going to OpenAI, Google, and xAI. The message to the industry is unmistakable: refuse the Pentagon's terms on how your model may be used, and you lose the Pentagon's business entirely.
And quietly, the episode revealed something the department did not volunteer: that even after being branded a national security risk, Anthropic's AI was still helping run American operations — including against Iran — right up until the end. The plug is now pulled, at least officially. Whether the replacement is safer, or just more compliant, is a question no official statement has answered.
Post a Comment