Pentagon expands classified AI roster

The Pentagon is moving military AI from pilot mode into infrastructure. That is the real story here. Instead of treating one chatbot or one model as the answer, the Defense Department is building a classified platform where multiple commercial AI systems can run inside its own secure environments. The immediate news is that seven more vendors were added for classified use through GenAI.mil, the department’s internal AI platform, after it had already started with Google’s model lineup. GenAI.mil has also gotten big fast — more than 1.3 million personnel have used it in five months, generating tens of millions of prompts and deploying hundreds of thousands of agents. (msn.com) ### What actually changed? The change is not just “the Pentagon signed some AI deals.” The Pentagon widened the roster of companies whose models and tools can be deployed on classified networks, specifically the IL6 and IL7 environments used for secret and other highly sensitive work. That turns GenAI.mil into a shared delivery layer for classified AI rather than a single-vendor product. (defensescoop.com) ### Which companies are in? Public reporting points to a roster that includes Google, OpenAI, xAI, Anthropic, Microsoft, Amazon Web Services, Nvidia, and Oracle — though some coverage has been inconsistent on whether the headline count is seven additional firms or eight total participants including Google’s earlier role. The importan(defensescoop.com)e stack on one supplier. (msn.com) ### Why does GenAI.mil matter so much? Because it already looks less like a demo and more like a workplace operating system. Five months in, more than 1.3 million Defense Department personnel had used the platform, producing tens of millions of prompts and hundreds of thousands of A(msn.com)ic infrastructure. (defensescoop.com) ### What are IL6 and IL7? They are the Pentagon’s security tiers for cloud and network environments. IL6 covers classified national security systems and secret data. IL7 goes further for highly sensitive classified workloads. So when officials say these models are being integrated into IL6 and IL7, they mean commercial AI is being pulled inside the Pentagon’s serious, mission-grade classified plumbing — not left on a public-facing sandbox. (defensescoop.com) ### Why use many vendors instead of one? Basically — leverage, resilience, and fit. Different models are good at different jobs. Some are better at coding, some at summarization, some at agents, some at retrieval-heavy tasks. A multi-vendor setup also reduces lock-in and gives the Pentagon bargaining power if one company stumbles on (defensescoop.com)ndor lock. (defenseone.com) ### What about the policy guardrails? Those have not gone away. The Pentagon’s broader responsible-AI framework still stresses lawful and ethical use, human judgment, and trust in outputs. Older and newer department guidance both keep the workflow human-centered — especially where AI outputs can affect operational or lethal decisions. That does not mean th(defenseone.com)drails that already exist. (media.defense.gov) ### What is the catch? The catch is that multi-model AI inside classified systems is hard to govern. You need tenant isolation, logging, access controls, and a clean abstraction layer so users can switch models without exposing sensitive data or rebuilding every workflow. The Pentagon is not just buying models. It is buying the cont(media.defense.gov)ry here — and the hard part. (defensescoop.com) ### Bottom line? This is the Pentagon saying AI is now a platform decision, not a pilot project. The vendors matter, but the bigger shift is structural: classified military AI is becoming a managed, multi-company utility embedded in everyday defense work. (msn.com)