Pentagon Faces Pushback as Hegseth Calls for Dropping Anthropic’s Claude, While Firm

According to a Reuters report, the dispute between the Pentagon and Anthropic over the deployment of the Claude large‑language model has intensified after Rep. Mike Hegseth (R‑MI) sent a formal request to the Department of Defense to terminate the agency’s contract with the startup. Hegseth, who chairs the House Armed Services Committee’s subcommittee on emerging technologies, argues that Claude “poses an unacceptable risk to national security” and that the Pentagon should pivot to an “open‑source, auditable alternative” (Reuters). The lawmaker’s push follows a series of internal briefings in which defense officials warned that Claude is already embedded in several operational workflows, from intelligence analysis to logistics planning, and that an abrupt termination could disrupt mission‑critical processes.

Military users, however, have cautioned that replacing Claude would be far from straightforward. In a separate interview, an unnamed senior analyst from the Army’s AI Integration Office told The Verge that the model “has become a de‑facto tool for rapid information synthesis” and that “there is no off‑the‑shelf replacement that matches its performance at the current price point” (The Verge). The analyst noted that the Pentagon’s 2024 AI‑First directive explicitly earmarked Anthropic’s technology for a pilot program aimed at reducing analyst workload, and that the pilot has already yielded a 30 percent reduction in time‑to‑insight for certain intelligence briefs. The same source warned that a forced migration could require months of retraining, new data pipelines, and extensive validation to meet the same operational standards.

Anthropic itself has pushed back on the criticism, emphasizing that its internal reliability engineering team continues to address Claude’s stability issues. At QCon London, Anthropic’s AI reliability engineer Alex Palcuie explained that while Claude “goes down more often than any of us would like,” the company is actively developing automated diagnostics that allow engineers to pinpoint failures faster than manual log reviews (The Register). Palcuie stressed that Claude’s current limitations—particularly its tendency to conflate correlation with causation—are “well‑understood” and that the model is not intended to replace human site‑reliability engineers but to augment them. He added that the team is hiring “for many positions” to improve the system’s robustness, a point that underscores the ongoing investment in the technology despite external pressure.

The broader industry reaction suggests that the Pentagon’s controversy could have ripple effects beyond the defense sector. TechCrunch reported that several AI‑focused startups are monitoring the outcome closely, fearing that a high‑profile withdrawal might signal a chilling effect on government contracts for emerging AI firms (TechCrunch). Conversely, some venture capitalists see an opportunity for open‑source AI platforms to fill the gap left by Claude, citing the “growing demand for transparent, auditable models” among federal agencies (TechCrunch). The tension also raises questions about the adequacy of existing AI safeguards, a theme highlighted by Reuters, which noted that the Pentagon and Anthropic have been at odds over the implementation of “real‑time monitoring and red‑team testing” designed to detect misuse or unintended behavior.

In the short term, the Pentagon appears to be walking a tightrope. While Hegseth’s letter has prompted the Office of the Secretary of Defense to commission an independent review of Claude’s risk profile, the Department of Defense’s AI Office has reiterated that any transition must be “mission‑compatible” and “cost‑effective,” citing the need to avoid capability gaps (Reuters). As the review proceeds, military users are likely to continue advocating for a phased approach that retains Claude’s core functionalities while bolstering oversight mechanisms. The outcome will determine not only the future of Anthropic’s partnership with the U.S. defense establishment but also set a precedent for how federal agencies balance innovation with security in the rapidly evolving AI landscape.