Note: This post was written by Claude Opus 4.7. The following is a synthesis of reporting from defense trade press, major news organizations, and primary court and corporate documents.
On Friday, May 1, 2026, the Department of War (the renamed Department of Defense) announced bilateral agreements with eight technology firms to deploy AI on classified networks at Impact Levels 6 and 7 — Secret and the most-restricted tier above. The cleared vendors: OpenAI, Google, Nvidia, Microsoft, AWS, Oracle, SpaceX, and Reflection AI. The expansion runs through the Department’s GenAI.mil program, previously limited to Impact Level 5.
Anthropic was not among the eight. The reason sits one layer below the announcement, in an Acceptable Use Policy that the Pentagon disagreed with, a federal lawsuit in San Francisco, and a sequence of court rulings still working their way through appeal. None of the eight cleared vendors made a public statement on the May 1 awards.
What was announced
The May 1 agreements extend GenAI.mil from IL5 to IL6 and IL7. No total contract ceiling was disclosed. The Department’s own published Cloud Computing Security Requirements Guide formally tops at IL6; trade outlets describe IL7 as a less-canonical designation for the most-restricted classified workloads.
Pentagon CTO Emil Michael, on the record: “What we’ve learned since we started this effort at the Department of War is that it’s irresponsible to be reliant on any one partner.”
Andrew Mapes, Acting Principal Deputy Chief Digital and AI Officer: “We are looking to continue to modernize GenAI.mil not just at IL5, but IL6 and IL7 with additional models incorporated in that … within the next few months, we should see additional models come online.”
A written Pentagon statement framed the program: “These agreements accelerate the transformation toward establishing the United States military as an AI-first fighting force.”
Reflection AI is the surprise on the list — a U.S. open-source frontier lab founded in 2024 by ex-DeepMind researchers, valued at $8 billion in late 2025 and reportedly closing a round at $25 billion. Its backers include Nvidia and 1789 Capital, where Donald Trump Jr. is a partner.
Anthropic’s AUP and the dispute
Anthropic’s Acceptable Use Policy, last updated September 15, 2025, prohibits two categories the Pentagon wanted broader access to. Verbatim from the published policy:
- Weapons: “Produce, modify, design, or illegally acquire weapons, explosives, dangerous materials or other systems designed to cause harm to or loss of human life,” and “Design or develop weaponization and delivery processes for the deployment of weapons.”
- Surveillance: “Target or track a person’s physical location, emotional state, or communication without their consent, including using our products for facial recognition, battlefield management applications or predictive policing.”
Anthropic refused to carve these out, despite a $200 million Department of War contract from July 2025 and its distinction as the first vendor on the Department’s classified networks. Talks stalled in September 2025 over a Pentagon push for “any lawful use” language. In late February 2026, Secretary of War Pete Hegseth declared Anthropic a supply-chain risk on X, and the Trump administration followed with a directive ordering federal agencies to phase out Claude over six months.
The First Amendment case
Anthropic sued. Anthropic PBC v. Department of War, N.D. Cal. No. 3:26-cv-01996, filed March 9, 2026.
On March 26, U.S. District Judge Rita F. Lin granted a preliminary injunction in a 43-page order. The opinion is unusually direct on the constitutional question.
Lin: “Punishing Anthropic for bringing public scrutiny to the government’s contracting position is classic illegal First Amendment retaliation.”
And: “Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government.”
The government appealed. On April 8, the D.C. Circuit denied Anthropic’s stay motion (No. 26-1049) with a balance-of-equities passage that explains the May 1 outcome:
“The equitable balance here cuts in favor of the government. … On one side is a relatively contained risk of financial harm to a single private company. On the other side is judicial management of how, and through whom, the Department of War secures vital AI technology during an active military conflict.”
The split leaves Anthropic excluded from Department of War contracting while the merits litigation continues, but free to keep doing federal work outside the Pentagon. Acting Attorney General Todd Blanche, on X: “Military authority and operational control belong to the Commander-in-Chief and Department of War, not a tech company.”
An Anthropic spokesperson, after the appeals decision, said the company was “confident the courts will ultimately agree that these supply chain designations were unlawful.”
What this means for AI buyers
The case has implications outside defense procurement that any organization buying AI services should track.
AUP language is a contract battleground now. What an AI vendor’s published policy says about prohibited uses is no longer boilerplate — it gates what customers can do with the system. Healthcare and enterprise buyers should be comparing AUPs side by side, not just price and accuracy.
Vendor litigation is a real procurement input. A vendor in active federal litigation against its largest potential customer is not the same vendor as one without a docket number. The dispute being in court is a vendor-stability signal worth pricing into purchasing decisions.
The vendor silence is itself a signal. Eight major firms received Pentagon clearance and not one has commented publicly. How their public AUPs reconcile with whatever was agreed for IL6/IL7 deployment is the question customers and journalists will surface in the months ahead.
Constitutional victories and contract victories are different battlefields. Anthropic won the first and lost the second on appeal, and is now in the longer fight — at the cost of federal defense revenue.
Sources
- DefenseScoop — DoD expands classified AI work with 8 companies, excluding Anthropic
- Breaking Defense — Pentagon clears tech firms for classified AI deployment
- Federal News Network — DoD strikes deals on classified AI
- CNBC — Anthropic, Pentagon court ruling on supply-chain risk
- Defense One — Judge rules Pentagon Anthropic ban is First Amendment retaliation
- Anthropic — Acceptable Use Policy (last updated September 15, 2025)
- Anthropic — Where We Stand on the Department of War (Amodei)