The United States Department of Defense announced this week that it has completely ceased using Anthropic’s AI products, ending a partnership that began in 2024. The move comes after Secretary of Defense Pete Hegseth designated Anthropic a “national security supply‑chain risk” in February, a label normally reserved for firms based in hostile foreign nations.
Verified events
Official cessation of use: A senior Pentagon official confirmed that all Anthropic tools, including the Claude language model and its advanced Mythos variants, are no longer employed in any department‑wide system.
Supply‑chain risk designation: In February, Hegseth publicly labeled Anthropic a supply‑chain risk, citing concerns that the company’s safety guardrails could be removed to give the military unrestricted AI access.
Legal pushback: Anthropic has called the designation “unprecedented and unlawful” and is pursuing litigation to overturn it.
Continued use until recently: Multiple sources told the BBC that Claude was still being used for research, analysis, and intelligence gathering up until last week, even after the February deadline.
Context and analysis
The Pentagon’s decision highlights the growing tension between rapid AI adoption and national‑security safeguards. While Anthropic’s Claude was praised for its ability to sift through massive data sets—satellite imagery, drone footage, and other intelligence—the depth of its integration made removal difficult. As Georgetown’s Center for Security and Emerging Technology senior analyst Lauren Kahn explained, “once they become integrated it can be painful to remove them.” This integration was most evident in the Maven Smart System, a Palantir‑run platform that serves as the Pentagon’s primary hub for organizing intelligence.
Anthropic’s refusal to strip safety guardrails, citing fears of mass surveillance and autonomous weapons, clashed with the Department’s desire for unfettered AI capability. The company’s stance underscores a broader industry debate: how to balance powerful AI functionality with ethical constraints. Hegseth’s aggressive timeline—six months from the February designation, with a deadline set on 27 February, the day before the U.S. and Israel launched strikes against Iran—suggests an urgency to mitigate perceived risks.
In the wake of Anthropic’s exit, the Pentagon has turned to other major AI providers. Contracts with Google, xAI, and OpenAI have been signed, and OpenAI’s tools are reportedly seeing broader adoption across military departments. This diversification may reduce reliance on any single vendor, but it also raises questions about consistency in safety standards across platforms.
Implications for future AI procurement
- Supply‑chain vetting will tighten: The Anthropic case sets a precedent for labeling domestic AI firms as security risks, potentially leading to more frequent blacklists.
- Integration challenges: Agencies must consider the technical cost of extracting AI components from entrenched systems like Maven.
- Legal and policy frameworks: Ongoing litigation may shape how future designations are applied and contested.
Uncertainty
While the Pentagon has confirmed the cessation of Anthropic tools, the exact timeline for fully purging Claude from all legacy systems remains unclear. Additionally, the impact of Anthropic’s lawsuit on the broader classification of AI supply‑chain risks is still pending judicial review.
Overall, the episode illustrates the delicate balance the U.S. military must strike between leveraging cutting‑edge AI for strategic advantage and safeguarding against the very risks such technology can pose when unchecked.