Trump’s AI Memo Strips Vendors of Veto Power Over Military

President Trump signed a National Security Presidential Memorandum on Friday directing the U.S. military and intelligence community to accelerate AI adoption and explicitly prohibiting any commercial vendor from disabling or modifying an AI system that warfighters depend on without prior approval from the chain of command. The directive, NSPM-11, arrived while Anthropic, the first AI company to deploy its models on classified Pentagon networks, still has two active federal lawsuits contesting the Pentagon’s effort to blacklist it from government work.

The memo bans two practices at the center of that dispute: unlawful surveillance of American citizens and unauthorized use of AI in autonomous weapons decisions. Anthropic was designated a supply chain risk in February for refusing to lift its contractual restrictions on both.

What NSPM-11 Commands

The memorandum states that “no entity, commercial or otherwise” may disable, degrade, or modify an AI system that American warfighters depend on without prior approval. Michael Kratsios, director of the White House Office of Science and Technology Policy (OSTP), posted that language nearly verbatim on social media Friday, noting the directive “ensures no entity can disable or degrade an AI system our warfighters depend on without prior approval.”

Secretary of War Pete Hegseth has 90 days from Friday to issue an updated version of DOD Directive 3000.09, the Pentagon’s governing guidance on autonomy in weapon systems, last revised in January 2023. Annual reviews are required thereafter to keep the directive current with AI advances. NSPM-11 also mandates AI procurement from multiple vendors and rescinds the Biden administration’s NSM-25. The White House fact sheet on the memorandum says NSM-25 had “burdened American AI adoption with ideological mandates and fostered dangerous single-vendor dependencies.” Commanders, directors, and agency heads are held responsible for ensuring obligations are met at every level of command.

The memo arrived three days after Trump signed “Promoting Advanced Artificial Intelligence Innovation and Security,” a separate executive order establishing a voluntary framework for AI companies to submit their most capable models for government cybersecurity testing before public release. That order set a 30-day pre-release window; an earlier draft proposed 90 days before the administration pulled it over concerns it would slow U.S. labs’ competitiveness against China. Together, the two documents form the administration’s AI-and-security framework, with the June 2 order handling pre-release review and NSPM-11 establishing chain-of-command authority over systems once deployed.

• Rapid onboarding of AI from multiple vendors to eliminate single points of failure across military and intelligence systems
• Secretary of War must update DOD Directive 3000.09 on autonomous weapons within 90 days, then annually
• AI may not be used for mass domestic surveillance or to censor free speech
• No commercial entity may unilaterally disable or alter an AI system warfighters depend on
• New partnerships offered to private-sector companies willing to help secure military AI against foreign threats

The Deal That Became a Federal Case

The July 2025 Agreement

In July 2025, Anthropic signed a contract with the Pentagon under which its Claude model became the first frontier AI approved for use on classified government networks. The deal was part of a round of “frontier AI projects” worth up to $200 million each. The Pentagon accepted Anthropic’s acceptable use policy as part of the agreement, including restrictions on mass domestic surveillance of Americans and use of Claude in fully autonomous weapons without human oversight over targeting decisions.

The restrictions held for roughly seven months. By early 2026, the Pentagon was pushing to renegotiate, seeking the right to deploy Claude “for all lawful purposes” without further limitation. Anthropic held both red lines. Weeks of failed negotiations followed.

The Breakdown and Designation

At 5:01 p.m. on February 27, 2026, the Pentagon’s self-imposed deadline expired. Trump posted to Truth Social directing all federal agencies to immediately stop using Anthropic’s technology, with a six-month phase-out period for agencies including the Defense Department. Defense Secretary Hegseth posted to X the same afternoon that “no contractor, supplier, or partner” doing business with the military could “conduct any commercial activity with Anthropic.”

The Department of War delivered the formal supply chain risk designation to Anthropic by letter on March 3, 2026. It was the first such designation applied to an American company; the label had historically been reserved for entities connected to foreign adversaries, including Chinese hardware manufacturers and state-linked software vendors. The Pentagon cited two statutory authorities: 10 U.S.C. §3252, which grants the Defense Secretary independent power to exclude sources from national security procurements, and the Federal Acquisition Supply Chain Security Act of 2018 (FASCSA). Under FASCSA, implemented through FAR 52.204-30, defense contractors were required to certify they were not using Anthropic products in military-related work. The General Services Administration (GSA) also removed Anthropic from USAi.gov, the government’s centralized platform for agencies to test AI models.

Emil Michael, the War Department’s Undersecretary of Defense for Research and Engineering and chief technology officer, told CNBC the department couldn’t allow a company’s “policy preference that is baked into the model” to dictate what warfighters could do with their AI systems.

Anthropic filed two federal lawsuits six days after the designation, one in the Northern District of California and one in the D.C. Circuit, challenging the legal basis for the supply chain risk designation under the Administrative Procedures Act, the First Amendment, and due process. In its California complaint, Anthropic argued the Defense Department violated the APA by skipping mandatory procedures Congress required before excluding a vendor: a formal risk assessment, notification to the targeted company, an opportunity to respond, a written national security determination, and congressional notification. The company said in filings the designation could reduce its 2026 revenue by multiple billions of dollars and put hundreds of millions in private-sector contracts at risk. Dozens of researchers at OpenAI and Google DeepMind filed an amicus brief in their personal capacities supporting Anthropic, arguing the designation could harm U.S. AI competitiveness and discourage developers from publicly discussing the limitations of their own systems.

How the Pentagon Rebuilt Its AI Roster

The May Classified-Network Agreements

On May 1, 2026, the War Department announced classified-network agreements with eight AI companies, making the Anthropic exclusion administratively complete. Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, Oracle, Reflection (a startup backed by NVIDIA), and SpaceX agreed to deploy their capabilities on the Pentagon’s Impact Level 6 (IL6) and Impact Level 7 (IL7) environments; IL6 covers information classified to the secret level, IL7 handles highly restricted compartmented data. Within five months of its launch, over 1.3 million War Department personnel were active on GenAI.mil, the military’s enterprise generative AI platform, generating tens of millions of prompts and running hundreds of thousands of AI agents across the department, with officials reporting the platform was cutting many tasks that previously took months down to days.

OpenAI’s Red Lines

OpenAI published its full agreement terms the same day. Its contract with the War Department states the AI system “shall not be intentionally used for domestic surveillance of U.S. persons and nationals,” language consistent with the Fourth Amendment and the Foreign Intelligence Surveillance Act of 1978 (FISA). On autonomous weapons, OpenAI committed not to deploy its models on edge devices, which forecloses their use in autonomous lethal targeting systems. The company described the arrangement as a cloud-only deployment architecture in which it independently verifies restrictions through classifiers it operates, rather than relying on the department’s usage policies alone.

OpenAI asked the War Department to make the same agreement framework available to all AI labs. The department separately announced it would convene a working group with frontier AI labs, cloud providers, and its own policy and operational communities to address AI capabilities, privacy, and national security challenges on an ongoing basis. The official press release described the multi-vendor design as intended to “prevent AI vendor lock and ensure long-term flexibility for the Joint Force.”

Where the Litigation Stands

The California Injunction

In late March 2026, U.S. District Judge Rita Lin, a Biden appointee in the Northern District of California, granted Anthropic a preliminary injunction blocking enforcement of the supply chain designation. She ruled the governing statute doesn’t support the “Orwellian notion” that an American company could be branded a potential adversary for publicly disagreeing with government policy, finding the designation violated Anthropic’s First Amendment and due process rights. Judge Lin also acknowledged the financial dimension: the designation, if enforced, threatened to reduce Anthropic’s 2026 revenue by multiple billions of dollars and put hundreds of millions in private-sector contracts at risk. The government appealed.

The D.C. Circuit Case

The D.C. Circuit denied Anthropic’s request for a stay in April 2026, ruling that the company’s claimed injuries “seem primarily financial in nature.” Acting Attorney General Todd Blanche posted to X after the ruling:

Military authority and operational control belong to the Commander-in-Chief and Department of War, not a tech company.

At oral arguments in May 2026, the bench showed sharp division. Judge Karen LeCraft Henderson, nominated by President George H.W. Bush, said she saw “no evidence to support the Pentagon’s determination” and called the designation “a spectacular overreach by the Department.” Judge Neomi Rao, a Trump nominee, questioned whether any court had basis for second-guessing a Defense Secretary’s national security judgment.

NSPM-11 now sits in the record as executive doctrine. The six-month phase-out Trump ordered on February 27 is scheduled to expire around late August 2026; the California injunction currently blocks its enforcement, while the D.C. Circuit case proceeds on a separate timeline. Anthropic has said publicly that its suits reflect no desire to disengage from national security work; the company’s position in filings is that the government terminated the relationship without legal authority to do so.

Trump said Friday he planned to host AI executives “as soon as next week.” Anthropic did not immediately respond to a request for comment on the memorandum or the planned meeting.