Pentagon Memo Sets 180-Day Countdown to Pull Anthropic AI From Military Systems

The Pentagon has given the U.S. military 180 days to cut Anthropic out of its systems — but quietly left the door cracked open.

An internal memo dated March 6 and signed by Pentagon Chief Information Officer Kirsten Davies directs officials to prioritize removing Anthropic's products from critical systems, including those supporting nuclear weapons and ballistic missile defense. But it also creates an exemption pathway: units can seek continued use "in rare and extraordinary circumstances" for "mission-critical activities directly supporting national security operations where no viable alternative exists." Any exemption requires a comprehensive risk mitigation plan.

The memo gives Pentagon contracting officers 30 days to notify contractors, who must then certify full compliance by the 180-day mark. The Pentagon confirmed the document but declined to comment further. Anthropic did not immediately respond to a request for comment.

The carve-out is less a concession than a recognition of reality, said Franklin Turner, a government contracts lawyer at McCarter & English. "I do expect to see a flurry of waiver requests," he told Reuters. Contractors may find it difficult to certify their entire software stack is free of any open-source code originating from Anthropic — a company whose models are widely embedded in enterprise AI tooling.

Anthropic has filed suit against the Department of Defense, arguing the blacklisting violates its First Amendment rights. The company's position is that it agreed to nearly all of the DoD's use cases — roughly 98 to 99 percent, according to CEO Dario Amodei — but drew lines against two categories: domestic mass surveillance and fully autonomous lethal weapons without human oversight.

The lawsuit reveals the extent to which Anthropic has already been embedded in military operations. The company developed "Claude Gov," a government-specific variant that is "less prone to refuse requests that would be prohibited in the civilian context, such as using Claude for handling classified documents, military operations, or threat analysis." The Washington Post reported that military personnel access Claude through a classified system called Maven — the same program Google abandoned in 2019 after employee protests, and which Palantir subsequently took over.

According to reporting by The Guardian, the government has used Claude for target selection and analysis in its bombing campaign against Iran. Amodei has stated he does not believe Anthropic has any role in operational military decision-making.

The standoff has exposed a broader shift in big tech's relationship with the military. On the same day Defense Secretary Pete Hegseth declared Anthropic a supply chain risk, OpenAI secured a deal allowing its technology in classified military systems. OpenAI's chief product officer also serves as a lieutenant colonel in the U.S. military's "executive innovation corps." Google, which dropped its Project Maven drone footage contract following mass employee protests in 2018, announced this week it would provide its Gemini AI to the military for building AI agents on unclassified projects.

Anthropic CEO Dario Amodei has framed the conflict as a narrower disagreement than it appears. In a blog post, he wrote that "Anthropic has much more in common with the Department of War than we have differences." He has advocated for supplying democratic militaries with advanced AI to counter autocratic adversaries — a position that complicates the company's image as an AI safety holdout. His stated red lines are narrower than many assume: he has argued AI should be usable for national defense "in all ways except those which would make us more like our autocratic adversaries."

The 180-day clock is running. Whether Anthropic's exemptions hold up in practice — and whether the First Amendment suit succeeds — will determine how cleanly the company can maintain its stated principles, or how much ground it has already conceded.