Weaponizing AI? Pentagon’s Bold Pressure Move

A hand interacting with a digital tablet displaying AI graphics

The Pentagon is pressuring a private AI company to drop “guardrails” that could help prevent domestic surveillance and autonomous-weapons use—raising a hard question about who sets the limits in America’s next battlefield.

Quick Take

Defense Secretary Pete Hegseth delivered a deadline-driven ultimatum to Anthropic over military access to its Claude AI model.
Anthropic says it will support national security work, but it draws lines against mass surveillance of Americans and fully autonomous weapons development.
The Pentagon has signaled it may use contract leverage—and potentially the Defense Production Act—to force compliance, a novel application for AI.
Claude is described as central to sensitive Defense operations, meaning a breakdown could create near-term capability gaps.
The outcome could set a precedent for how much control Washington can exert over private AI safety policies in defense contracting.

Ultimatum Deadline Puts AI Control Fight on a Fast Clock

Defense Secretary Pete Hegseth set a Friday evening, Feb. 28, 2026 deadline after a Tuesday meeting with Anthropic CEO Dario Amodei, demanding “unrestricted” military access to Claude and warning of steep consequences if the company refused. Reporting indicates the Pentagon threatened contract termination and even a “supply chain risk” designation. As of the deadline date, the dispute’s resolution remained unclear in available public reporting.

Accounts of the meeting’s tone differ, with one Defense official calling it “not warm and fuzzy,” while another described it as cordial and said Hegseth praised Claude’s performance. The split descriptions matter because they hint at two competing realities: a negotiation where both sides know they need each other, and a pressure campaign designed to make an example out of a safety-focused firm that won’t fully yield on usage rules.

Anthropic’s “Red Lines” Focus on Americans’ Privacy and Autonomous Weapons

Anthropic has presented its position as conditional cooperation rather than refusal to help the military. The company has indicated it is open to modifying usage guidelines for Pentagon needs while holding firm against two categories: widespread surveillance of Americans and development of autonomous weaponry. Those boundaries are not just corporate branding; they map onto constitutional anxieties voters already have about unaccountable monitoring and the growth of government power through new technologies that few citizens can audit or even understand.

The immediate flashpoint reportedly involved how Claude was used through a defense contractor partnership, with a specific dispute over whether Anthropic raised concerns to Palantir about a Venezuela-related operation sometimes referenced as the “Maduro raid.” Amodei disputed the characterization that Anthropic formally objected, describing it instead as routine operational dialogue. That distinction is crucial because it goes to whether Anthropic tried to “veto” a mission—or merely flagged technical and policy risks inside normal contractor communications.

Government Leverage: Contracts, “Supply Chain” Labels, and the Defense Production Act

Hegseth’s central message, as reported, was that the Department of Defense will not allow a corporation to set conditions on operational decisions or object to use cases. The Pentagon’s leverage appears to be contractual and regulatory: ending Anthropic’s work, labeling the firm as a supply chain problem, and floating the Defense Production Act as a pressure tool. The DPA was used prominently during COVID-19 to speed production, but applying it to compel changes to AI model safeguards would be a new and contentious frontier.

Legal uncertainty sits in the background. A defense consultant cited in reporting suggested Anthropic could challenge a DPA move in court by arguing that bespoke software tied to classified use differs from mass-produced goods the DPA has historically prioritized. That uncertainty helps explain the brinkmanship: the Pentagon may be trying to win compliance without litigating, while Anthropic may be trying to preserve principles without triggering a full rupture that could jeopardize a major government relationship.

National Security Risk vs. Corporate Gatekeeping: A Precedent in the Making

The Pentagon’s own reported posture suggests it recognizes the practical risk of losing Claude. A Defense official acknowledged Anthropic’s leverage, saying the government is still talking because it needs them “now,” and describing the company as exceptionally capable. Reporting also states Claude has become the only AI model used for some of the military’s most sensitive operations. If that is accurate, swapping to a competitor on short notice could be difficult, especially for cyber or other specialized missions.

The longer-term stakes go beyond which vendor wins a contract. If the government can force a private model provider to remove safety limits as a condition of doing business, then “guardrails” become optional whenever federal power is applied hard enough. If, on the other hand, a contractor can hold firm against certain mission categories, the public will likely demand clearer lines: what is lawful, what is constitutional, and what should be off-limits even if it is technically feasible. Public reporting does not yet provide final terms, so the key takeaway is the precedent, not the outcome.