Pentagon Flags Anthropic’s Chinese Staff as Security Threats for US

Anthropic’s contract with the Defense Department, a $4 billion deal announced in early March, has now become a flashpoint for national‑security concerns. The Pentagon’s own assessment, reported by Axios, flags the company’s Chinese‑national employees as “potential security risks” because they could be subject to foreign intelligence influence or coercion. The internal review, part of the Department of Defense’s broader supply‑chain‑risk program, notes that Anthropic employs roughly 150 staff members who are Chinese citizens or permanent residents, many of whom work on the Claude chatbot’s core language‑model training pipelines. According to the Axios piece, the Pentagon’s risk matrix assigns these workers a “high‑impact” rating, meaning any compromise could affect classified or mission‑critical AI outputs used by the military.

The move follows a parallel legal battle in which the Justice Department, under the Trump administration, sued Anthropic for allegedly failing to safeguard “warfighting systems” from foreign interference. Wired reported that the department’s filing argued the company’s supply‑chain practices violated federal security standards, even as the administration claimed the designation did not infringe on Anthropic’s First Amendment rights. The lawsuit contends that Anthropic’s data‑handling procedures lack the “segregation and vetting” required for defense contracts, a claim that the company rebuts by pointing to its internal compliance program and the fact that its Chinese staff are subject to the same background checks as all employees.

Anthropic has pushed back hard. In a filing covered by Wired, the startup sued the Department of Defense, accusing the agency of overstepping its authority by turning a contractual dispute into a de‑facto ban on its technology. CEO Dario Amodei’s legal team argues that the Pentagon’s supply‑chain designation is “arbitrary” and “unsupported by concrete evidence of espionage,” noting that Anthropic already encrypts model weights and isolates training data from external networks. The company also emphasizes that its Chinese engineers are integral to research collaborations with U.S. universities and that none have been found to have undisclosed foreign affiliations, a point the defense‑risk report does not address.

TechCrunch’s coverage of the Pentagon’s designation adds another layer: the agency is now moving to place Anthropic on a formal “restricted‑technology” list, which would require the firm to obtain a special waiver for any future DoD work. The article explains that such a listing could halt the rollout of Claude‑based tools across the services, forcing the military to seek alternative vendors or develop in‑house models. Analysts cited by TechCrunch warn that the decision could ripple through the broader AI market, as other contractors may face heightened scrutiny over foreign‑national staff. If the Pentagon’s risk assessment holds, Anthropic could see a slowdown in its defense pipeline just as the company expands its commercial offerings, potentially reshaping the competitive landscape between U.S. AI firms and their overseas rivals.