Sen. Wyden warns of mass surveillance as Pentagon battles Anthropic over AI

The Pentagon’s push to embed Anthropic’s Claude model in defense systems has run into a legal and ethical impasse that could reshape how the Department of Defense acquires commercial AI, according to a Fortune briefing that outlines three core questions driving the dispute. First, the DoD must decide whether it can mandate the use of a private‑sector model without imposing its own data‑handling rules, a point Anthropic has resisted by demanding “the bare minimum ethical guardrails” on how the department would employ Claude. Second, the agency is grappling with whether the model can be cleared for fully autonomous weapons—a use case Anthropic explicitly refused in a public letter on Wednesday, citing concerns over unintended escalation. Third, the broader issue of data provenance looms large: the DoD’s reliance on third‑party data brokers to feed Claude raises questions about the legality of aggregating location, browsing, mental‑health, political and religious information that, as Senator Ron Wyden warned, is “available for pennies on the open market” (Gizmodo).

Wyden’s alarm is rooted in the potential for mass surveillance once the DoD can fuse disparate data streams into detailed profiles of U.S. citizens. He argues that the department’s “collection of data is happening through private data brokers,” and that the combination of such data with Claude’s generative capabilities could produce “highly revealing profiles of Americans” (Gizmodo). The senator, a long‑time privacy advocate, announced plans to introduce legislation that would restrict the DoD’s ability to purchase and merge commercial data without explicit congressional oversight. His proposal seeks to close the loophole that allows the government to sidestep existing privacy statutes by leveraging the commercial market’s low‑cost data, a concern echoed by privacy scholars who note that the line between legitimate national‑security analytics and unlawful surveillance is increasingly blurred.

The political fallout has already prompted an executive decision. Former President Donald Trump announced that the federal government would cease using Claude within six months, a move that Defense Secretary Pete Hegseth reinforced by stating that any contractor wishing to do business with the DoD must also disengage from Anthropic (Gizmodo). Anthropic, for its part, has signaled it will sue the department, arguing that the DoD’s demand for unrestricted access to Claude violates the company’s ethical commitments and could set a precedent for future government‑industry contracts. Legal experts anticipate a protracted battle, noting that the government’s leverage over a single AI vendor is limited, especially as the market expands with alternatives like OpenAI’s GPT‑4 and Google’s Gemini, which are already being integrated into defense pipelines.

Meanwhile, Anthropic is advancing its own data‑integration roadmap, unveiling a Model Context Protocol designed to standardize how AI systems connect to external datasets (VentureBeat). The protocol aims to give developers granular control over data provenance and usage policies, a feature that could address some of the DoD’s concerns about “ethical guardrails.” However, the protocol’s rollout does not resolve the immediate dispute over Claude’s deployment in classified environments, and the company’s recent launch of interactive Claude apps for platforms such as Slack (TechCrunch) underscores its broader commercial ambitions that may conflict with a narrow, defense‑only focus.

The clash illustrates a broader tension between the rapid militarization of generative AI and the need for robust governance frameworks. As the DoD seeks to harness AI for faster decision‑making and threat analysis, the lack of clear, enforceable standards for data ethics and privacy threatens to stall adoption and invite legislative pushback. Wyden’s proposed bill could become a template for future oversight of AI procurement, compelling the Pentagon to adopt stricter data‑handling protocols or risk losing access to cutting‑edge models. The outcome of this standoff will likely set the tone for how the United States balances national security imperatives with civil liberties in the age of AI‑driven surveillance.