Image: Flickr / Wikimedia Commons
The Pentagon's AI Roster Is Eight Companies Long - and Anthropic Isn't on It
The Defense Department has signed classified-network AI agreements with eight tech firms. The absence of one name tells the more important story.
This article was produced by the AETW editorial team.
The U.S. Department of Defense has signed classified-network AI agreements with eight companies — Nvidia, Microsoft, AWS, Google, OpenAI, SpaceX, Reflection AI, and Oracle — explicitly bypassing Anthropic, which is currently suing the Pentagon over its supply-chain risk designation.
Eight companies, two classification tiers
On May 1, 2026, the U.S. Defense Department announced it had signed agreements with eight frontier AI companies — Nvidia, Microsoft, Amazon Web Services, Google, OpenAI, SpaceX, Reflection AI, and Oracle — authorizing them to deploy their AI hardware and models on its classified networks for what the department described as 'lawful operational use.'
The deployments will operate across two security tiers: Impact Level 6, which covers classified data up to the Secret level, and Impact Level 7, the most stringent classification covering compartmented intelligence and the most sensitive operational systems. Both tiers require physical security controls, strict access restrictions, and continuous auditing. This is where the U.S. military runs actual war-fighting decision support — not administrative back-office work.
The Anthropic absence
Anthropic is not on the list. That is the most consequential detail in the announcement. The company's Claude models were already in use on classified networks via Palantir's Maven toolkit, and the NSA has reportedly been using Anthropic's not-yet-public Mythos model for its cyber warfare capabilities. The exclusion is not a technical disqualification — it is a procurement consequence of an ongoing legal dispute.
The dispute traces back to early 2026, when the Pentagon sought unrestricted use of Anthropic's AI tools. Anthropic refused, insisting on guardrails against domestic mass surveillance and autonomous weapons. The Defense Department responded by branding Anthropic a 'supply-chain risk' — a designation normally reserved for foreign adversaries. Anthropic sued. In March 2026, it won an injunction blocking that designation. The litigation is still active, but Friday's announcement makes clear that the courts can halt a label while doing nothing to reverse a procurement decision. The Pentagon simply signed deals with everyone else.
What Reflection AI is doing on this list
The most unexpected name in the agreement is Reflection AI. The startup was founded in March 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou. As of May 2026, it has not publicly released the open-weight frontier model at the center of its pitch. Its code research agent Asimov is still in waitlist status. Earlier this year, the Financial Times reported Reflection was raising at a $20 billion valuation despite shipping no public product.
The Pentagon deal is the first time a U.S. agency has contracted with Reflection at the operational tier. The company is backed by Nvidia and is pursuing what has been described as a deliberately American alternative to DeepSeek — an open-weight strategy designed to give the U.S. government a non-proprietary option. Inclusion on this list, before any public model ships, is a significant implicit endorsement of both the team and that strategic direction.
The scale already in motion
The classified-network agreements build on infrastructure the Pentagon has already been scaling rapidly. More than 1.3 million Defense Department personnel have used GenAI.mil — the department's secure enterprise generative AI platform — generating tens of millions of prompts across government-approved cloud environments. GenAI.mil is designed for unclassified tasks: research, document drafting, data analysis. The new IL6 and IL7 agreements extend AI access into the classified tier where operational decisions actually get made.
The Pentagon's CTO, Emil Michael, stated on Friday that the department concluded it was 'irresponsible to be reliant on any one partner' — a direct reference to the Anthropic situation. The stated goal is now an architecture that prevents AI vendor lock-in and maintains flexibility for what officials are calling an 'AI-first fighting force.' The eight-company roster reflects that: it spans proprietary frontier model labs, sovereign cloud infrastructure, GPU compute, and a satellite-internet operator.
The question the injunction cannot answer
Anthropic's legal position remains intact. The injunction blocking the supply-chain risk designation still stands, and the case will return to court. Anthropic's CFO previously warned that the designation could cost the company multiple billions of dollars in 2026 revenue. The injunction blunted that exposure — but it cannot compel procurement decisions. The Pentagon can, and did, simply contract with other providers instead.
The deeper issue is whether the guardrails Anthropic insisted on — restrictions on autonomous weapons use and domestic mass surveillance — were ever going to be compatible with how the Defense Department intended to operate. That tension has not been resolved by the lawsuit. It has been deferred. The Pentagon's eight-company roster is now operational proof that it believes it can build the AI-first military it wants without Anthropic in the room.
Sources
AI & Technology Researcher
Brian Weerasinghe is the founder and editor of AI Eating The World, where he covers artificial intelligence, tech companies, layoffs, startups, and the future of work. His reporting focuses on how AI is transforming businesses, products, and the global workforce. He writes about major developments across the AI industry, from enterprise adoption and funding trends to the real-world impact of automation and emerging technologies.
