Pentagon seals classified AI deals

The Pentagon just moved commercial AI a lot deeper into military systems. Not into a public chatbot sandbox — into classified network environments where warfighting, intelligence, and internal operations actually run. That matters because the hard part of military AI was never just getting a model to answer questions. It was getting outside tools cleared, contained, and usable inside locked-down systems. On May 1, the Defense Department said seven companies had signed agreements to do exactly that, with Oracle added later the same day. (nextgov.com) ### Which companies got in? The initial list was Google, Microsoft, Amazon Web Services, Nvidia, OpenAI, SpaceX, and Reflection. Later reporting said Oracle also joined, bringing the total to eight. That detail matters because the story is not “the Pentagon picked a champion.” Basically, it pi(nextgov.com)e vendor. (nextgov.com) ### What are IL6 and IL7? These deals cover Impact Level 6 and Impact Level 7 — the Defense Department’s classified cloud environments. IL6 handles data up to the secret level. IL7 is tighter and supports highly restricted, top-secret-adjacent national security workloads. So this is not the sa(nextgov.com) digital spaces. (federalnewsnetwork.com) ### What does the Pentagon want the AI to do? The Pentagon’s language is broad but revealing. It says the tools will help with data synthesis, situational understanding, and decision support in complex operations. In plain English, think less “robot generals” and more systems that c(federalnewsnetwork.com) commanders. The department also tied the effort to warfighting, intelligence, and enterprise operations, which means the use cases likely range from battlefield analysis to back-office workflows. (nextgov.com) ### Why is this happening now? Because the Pentagon already built a front door for this. Its GenAI.mil platform launched in December, and officials have been signaling that more models would be added across higher classification levels. Google’s Gemini was already available for sensitive but u(nextgov.com)e have a platform” to “we can actually populate it with multiple vendors inside classified environments.” (nextgov.com) ### Why does the multi-vendor part matter so much? Vendor lock-in is the catch. If one company becomes the only model provider inside classified systems, the Pentagon loses leverage on price, performance, and policy. The department said outright that the goal is to prevent lock-in and preserve(nextgov.com)nce is part of the architecture. (nextgov.com) ### Why is Anthropic missing? That absence is not random. Multiple reports tie it to a dispute earlier this year over ethical limits on military and surveillance uses. The Pentagon’s new lineup lands right after that fight, which makes the message pretty clear — if a vendor wants access to classified defense workflows, the department wants terms that allow “lawful operational use” on its side. (nextgov.com) ### So what actually changed? Before this, “military AI” often meant pilots, prototypes, or unclassified experimentation. Now the Pentagon is pushing commercial frontier models into the networks that matter most. The bottom line is simple: the bottleneck has shifted. The question is no longer (nextgov.com)assified stack. (nextgov.com)