Pentagon awards classified-network AI deals to OpenAI, Google and Microsoft — Anthropic excluded

The Pentagon just moved commercial AI one layer deeper into the military stack. Not into a sandbox, and not just for office productivity, but into classified network environments where the Defense Department runs sensitive operational work. That is the real news here. On May 1, the department said OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, SpaceX and Reflection will be allowed to deploy frontier AI capabilities on its classified systems for “lawful operational use.” ### What actually changed? These are formal agreements to bring outside AI tools into Impact Level 6 and Impact Level 7 environments. In Pentagon language, that means the systems approved for classified and top-secret style cloud workloads — the places with the hardest security requirements and the fewest vendors. The department says the tools will support the GenAI.mil platform. ### Why do IL6 and IL7 matter? Because this is the difference between “we use chatbots at work” and “your model can live near some of the government’s most sensitive data.” IL6 is the baseline for classified cloud processing. IL7 goes further for highly sensitive national-security workloads. Getting into those zones is hard — security reviews, isolation in one of the most locked-down computing environments in the country. ### Why this group of companies? The Pentagon is very plainly trying to avoid vendor lock-in. Its message is basically: no single lab gets to become the military’s operating system for AI. That is why the roster spans model companies, cloud providers, chip infrastructure and even SpaceX. The department said the point is to build an architecture with long-term flexibility and a “diverse suite” of capabilities across the American tech stack. ### So why is Anthropic missing? That is the sharp edge of the story. Anthropic is not just absent by accident — it has been in an open fight with the Pentagon over how far military use of its models should go. Multiple reports say the dispute centered on Anthropic refusing unrestricted use for autonomous weapons and domestic surveillance scenarios. The Defense Department then labeled the company a supply-chain risk, and the clash spilled into court. ### Is the exclusion permanent? Maybe not. There are signs the standoff could soften. Military Times reported White House officials met Anthropic CEO Dario Amodei on April 17, and President Donald Trump later said a deal was “possible.” But that is very different from being in the room now, while rivals are getting embedded into the Pentagon’s classified AI plumbing. ### Why does GenAI.mil matter here? Because the Pentagon already has a front door for AI adoption. TechCrunch reported more than 1.3 million Defense Department personnel have used GenAI.mil so far, mostly for approved enterprise tasks. These new agreements extend that idea upward — from safer internal use cases toward classified operational environments. Basically, the Pentagon is taking the tooling people already know and pushing it closer to mission systems. ### What is the bigger play? This fits the Pentagon’s broader push to become, in its own words, an “AI-first” fighting force. That phrase can sound like branding, but the concrete point is simpler: the department wants commercial AI inside defense workflows fast, and it wants enough suppliers that no one company can dictate terms later. Anthropic’s exclusion shows the price of resisting that push. Everyone else on the list just got a head start.