Pentagon approves 8 vendors for classified AI model network — Anthropic excluded
- The Pentagon approved eight companies to put frontier AI on classified military networks on May 1 — and Anthropic was not one of them. (defensescoop.com) - The cleared group is AWS, Google, Microsoft, NVIDIA, OpenAI, Oracle, Reflection, and SpaceX for IL6 and IL7 systems — basically secret and top-secret environments. (defensescoop.com) - What changed is the buying model. The Pentagon is moving from dependence on one lab toward a multi-vendor stack after its Anthropic fight. (defenseone.com)
The story here is military AI procurement — not just another flashy model launch. The Pentagon has now approved eight companies to deploy frontier AI on its classified networks, which means the tools can move closer to real defense workflows instead of staying in demos or unclassified sandboxes. (defensescoop.com) The part getting attention is who made the list and who did not. Anthropic, which had been deeply embedded in some Pentagon AI work, was left out on May 1. ### What actually got approved? The approved companies are Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, Oracle, Reflection, and SpaceX. The Pentagon said these agreements cover deployment into Impact Level 6 and Impact Level 7 environments — the classified cloud tiers used for secret and top-secret or similarly sensitive national security workloads. (defenseone.com) The tools are being funneled through GenAI.mil, the department’s central AI platform. ### Why do IL6 and IL7 matter? Because this is the line between “interesting software” and “operational system.” Plenty of companies can show a chatbot on a clean laptop. Far fewer can get software into the locked-down environments where intelligence, planning, cyber operations, and sensitive command support actually happen. (defensescoop.com) The Pentagon framed the move as a way to speed up data synthesis, improve situational awareness, and help warfighters make decisions faster. ### Why is Anthropic missing? That goes back to the Pentagon’s blowup with Anthropic earlier this year. The dispute centered on military-use restrictions Anthropic wanted to preserve around things like autonomous weapons and surveillance. (defensescoop.com) After that fight escalated in late February and early March, the Pentagon labeled Anthropic a supply-chain risk, and the relationship has not recovered. Even now, Pentagon CTO Emil Michael is still publicly calling Anthropic a supply-chain risk. ### Was Anthropic important before this? Yes — that is what makes the exclusion more than symbolic. Anthropic’s Claude had already been used on classified networks through Palantir’s Maven toolkit, so this is not a case where the Pentagon simply picked vendors it already trusted and ignored everyone else. (defensescoop.com) A company that had working access got pushed out while rivals were invited in. That is a real shift in leverage. ### Why approve so many vendors at once? The Pentagon’s own answer is vendor lock-in. Officials said the broad list was intentional because relying on one model provider is “never a good thing,” and the department wants a domestic ecosystem of model developers rather than a single chokepoint. (defenseone.com) In plain English — if one company balks, breaks, raises prices, or falls behind, the Pentagon wants substitutes already inside the fence. ### Why are names like Reflection and SpaceX on the list? Because this is not just about the biggest chatbots. It is about assembling a stack — models, cloud infrastructure, compute, and mission-specific integration. Reflection is a newer startup, NVIDIA brings the hardware and AI platform layer, AWS and Oracle bring cloud plumbing, and SpaceX suggests the Pentagon wants options that connect AI with its broader defense-tech ecosystem. (breakingdefense.com) That mix tells you the department is buying capability breadth, not a single winner. ### Does this settle the Anthropic fight? Not really. There is still legal and policy spillover from the earlier blacklist fight, and there are signs parts of government remain interested in Anthropic’s newer Mythos model for cyber use cases. (defenseone.com) But inside this specific classified-network expansion, the message is blunt — the Pentagon is building around Anthropic, not waiting for a reconciliation. ### Bottom line? The Pentagon just turned classified AI access into a competitive market. That is good news for OpenAI, Google, Microsoft, AWS, NVIDIA, Oracle, SpaceX, and Reflection. And it is a warning to every frontier lab that wants defense business — if your usage terms clash with the Pentagon’s, the department now has plenty of other doors to walk through. (defensescoop.com) (defenseone.com) (cnbc.com)