This report chronicles the dispute between Anthropic and the US Department of Defense over use restrictions on Anthropic's models, spanning the June 2025 launch of Claude Gov, the January 2026 allegation that Claude was used in a Venezuela operation, Defense Secretary Pete Hegseth's February 27, 2026 ultimatum, an accompanying Trump directive to federal agencies, and a March 26, 2026 preliminary injunction issued by Judge Rita Lin.
Claude Gov
Anthropic released Claude Gov on June 6, 2025, for US national security customers, describing it as already deployed at the "highest level of U.S. national security." The product offered four enhancements over standard Claude: classified-material handling, intelligence analysis, linguistic expertise, and cybersecurity intelligence. Anthropic stated that Claude Gov underwent the same safety testing as its other Claude models. The public sector contact listed was pubsec@anthropic.com.
Use restrictions and the Pentagon dispute
The dispute originated in a disagreement over the terms governing Anthropic's models in government use. Anthropic's Dario Amodei maintained two restrictions he described as non-negotiable: no fully autonomous weapons systems, and no mass surveillance of American citizens. The source notes these align with the ICRC Position on Autonomous Weapon Systems (source summary) and DoD Directive 3000.09 — Autonomy in Weapon Systems (source summary), with Anthropic taking a stricter posture on autonomy than the DoD's own directive.
In January 2026, the US military allegedly used Claude in the Venezuela/Maduro operation, as reported by Reuters and Axios; Anthropic denied discussing specific operations with the Department of War (DOW). In January and February 2026, Amodei objected to the Pentagon's push to renegotiate contracts to permit "all lawful use," reiterating the two restrictions. Axios reported on February 15, 2026 that the Pentagon was threatening to cut off Anthropic. Hegseth set a February 27, 2026 deadline for Anthropic to comply; Anthropic did not comply. That same day, a Trump directive instructed federal agencies to cease using Anthropic's products, and Hegseth designated Anthropic a supply chain risk. Military Times reported on March 19, 2026 that military users described dumping Claude as "not so easy."
The Pentagon's chief technology officer characterized lab-level use restrictions binding the government as "not democratic," framing the underlying question of who decides how high-capability AI is used.
Preliminary injunction
On March 26, 2026, Judge Rita Lin issued a preliminary injunction in a 43-page ruling blocking the government ban and the supply-chain-risk designation, finding the retaliation likely illegal. The source frames the ruling as a First Amendment and administrative-law matter, holding that retaliatory government action against a vendor for its public-policy positions is enjoinable. It also characterizes the episode as the first public clash between a frontier AI lab and the DoD over use restrictions, and as a precedent for whether frontier-lab use policies bind the US government as a customer.
Related events
The source situates the dispute against the Anthropic $1 OneGov deal with the GSA (August 12, 2025), reached roughly six months before the Pentagon dispute. It cites Dean Ball's "Clawed" analysis (Clawed) as commentary on the policy tensions, and the NYU Stern Center for Business and Human Rights "The Cost of Conscience" quick-take.
Relationships
- supports: Autonomous Weapons, ICRC Position on Autonomous Weapon Systems (source summary), AI and National Security, AI Whistleblowing (Amodei's public refusal as protected speech).
- contradicts: Pentagon's "all lawful use" posture.
- depends-on: Anthropic, Dario Amodei, DoD Directive 3000.09 — Autonomy in Weapon Systems (source summary).
- related: Clawed, (Source: nytimes.com), GSA OneGov Program and USAi Platform (August 2025), Palantir Technologies, Anduril Industries.