Pentagon inks classified‑AI access deals with seven firms

The Pentagon just moved commercial AI one step deeper into its most sensitive systems. On May 1, the Defense Department said seven companies will be allowed to bring AI tools into classified military networks. That matters because this is not the usual “pilot program in a sandbox” story. These are Impact Level 6 and 7 environments — the part of the stack built for secret and highly restricted national security work. ### Which companies got in? The initial list was Google, Microsoft, Amazon Web Services, Nvidia, OpenAI, SpaceX and Reflection. Reflection is the odd one out for most readers — a newer AI startup backed by Nvidia, not a giant cloud incumbent. The department said these companies will support classified workflows through its GenAI.mil platform, which is becoming the Pentagon’s central front door for generative AI. ### What are IL6 and IL7, exactly? They are the Pentagon’s security tiers for cloud and data environments. IL6 is used for information classified up to secret. IL7 is the tighter zone for highly restricted, more sensitive national security workloads. So this announcement is really about moving frontier AI from “useful office assistant” territory into places tied to operational and intelligence work. ### What does the Pentagon want these models to do? The short version is faster synthesis and faster decisions. The department says the tools are meant to speed up data analysis, improve situational awareness and help people make decisions in complicated operational settings. That can mean battlefield is framing all of this as part of an “AI-first fighting force.” ### Why sign so many vendors at once? Basically — they do not want to get trapped with one provider. The Pentagon’s own language stressed avoiding vendor lock-in and keeping long-term flexibility for the joint force. That is a procurement choice, but it is also a power move. If one lab balks at terms, the department wants alternatives ready on the shelf. ### So why is Anthropic missing? Because this is also a contract fight in disguise. Anthropic had been battling the Pentagon over guardrails, especially limits on use in fully autonomous weapons and surveillance of Americans. That dispute spilled into court after the Trump administration pushed agencies to stop like the Pentagon building around a resistant supplier. ### Wait — is it seven firms or eight? Turns out the count changed the same day. The Pentagon’s first announcement named seven companies, but later reporting said Oracle also agreed to join, bringing the total to eight. That means the headline number depends on which version of the day’s paperwork you saw first. The broader point does not change — the supplier pool for classified military AI just got wider. ### Is this already live? At least part of it is. Officials said military personnel, civilians and contractors are already using GenAI.mil, and more models are expected to come online over the next few months across different classification levels. The catch is that public reporting still leaves one path. ### Bottom line This is less about one contract than about the Pentagon setting the market structure for military AI. It wants multiple American vendors, available on classified networks, under government terms — and it is willing to sideline companies that try to narrow how those tools can be used.