Pentagon signs deals, cuts Anthropic

The Pentagon just made a very specific kind of AI bet. Not a flashy chatbot bet. A classified infrastructure bet. On May 1, it said seven companies — Google, Microsoft, Amazon Web Services, Nvidia, OpenAI, SpaceX, and Reflection — can bring AI capabilities into its classified environments, including Impact Level 6 and 7 networks used for secret and top-secret work. (nextgov.com) ### What actually changed? The new agreements widen the bench of companies allowed to support classified Pentagon workflows through GenAI.mil, the department’s internal AI platform. The Pentagon said the goal is faster data synthesis, better situational awareness, and quick(nextgov.com) ### Why do IL6 and IL7 matter? Because this is the hard part. Lots of AI tools can demo well on an unclassified laptop. Far fewer can run inside the Pentagon’s most sensitive cloud environments. IL6 is the compliance bar for classified defense cloud workloads. IL7 goes furt(nextgov.com) it can survive the security, hosting, and integration gauntlet that real defense work demands. (defensescoop.com) ### Why is Anthropic missing? Because the fight stopped being only about model quality. Anthropic pushed for contract limits that would block use of its models in fully autonomous weapons and in surveillance of Americans. The Pentagon rejected that approach, saying lawful uses had to remain a(defensescoop.com)tracts, even as court fights over the broader federal ban kept going. (abcnews.com) ### Is this really about “supply chain risk”? Partly, yes — but basically that label is doing two jobs at once. It is a formal security designation, and it is also a procurement weapon. Once Anthropic got tagged, the Pentagon could move faster to replace it with vendors willing to accept the department’s operatin(abcnews.com) the surrounding cloud, compute, and delivery machinery too. That last point is an inference from the vendor mix and the deployment scope. (nextgov.com) ### Why these companies? Look at the list and the pattern jumps out. Microsoft and AWS bring cleared cloud infrastructure. Nvidia brings the chip layer. Google and OpenAI bring frontier models. SpaceX and Reflection bring more specialized operational software and integration paths. The Pentagon is buying a menu of interoperable pieces for classified use, not crowning one “best model” winner. (nextgov.com) ### What does this mean for the AI market? It raises the value of being deployable, not just impressive. In commercial AI, leaderboard performance gets attention. In defense AI, the bigger moat may be cleared infrastructure, security paperwork, and the ability to plug into (nextgov.com)ise on terms or partner more deeply with infrastructure players. (nextgov.com) ### Is this already live? Yes. Pentagon officials said personnel are already using AI capabilities through GenAI.mil, and one official said the department’s use of Google’s models has already been saving thousands of man-hours each week. So this is not a pilot in the abstract. The plumbing is already being used. (abcnews.com) ### The bottom line? The real story is not just that Anthropic got cut. It is that the Pentagon is deciding who gets to be part of the classified AI stack — and that decision now rewards security clearance, infrastructure depth, and willingness to operate on the government’s terms. (nextgov.com)413264/))