Pentagon expands classified AI deals

The Pentagon just made its AI push much more concrete. On May 1, the Defense Department said eight companies will now help deploy frontier AI tools on its classified networks — the systems used for highly sensitive military work. That matters because this is no longer about office copilots or unclassified pilots. It is about putting commercial AI inside Impact Level 6 and Impact Level 7 environments, where the military handles classified and top-secret workloads. ### What actually changed? The new agreements cover OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, Oracle, SpaceX, and Reflection. The Pentagon said these companies will provide resources to deploy their capabilities for “lawful operational use” on classified networks. The stated goal is pretty direct — faster data synthesis, better situational awareness, and stronger decision support for warfighters. ### Why do IL6 and IL7 matter? Because those labels are the line between ordinary government AI use and the hard stuff. IL6 is the Defense Department’s cloud standard for classified workloads. IL7 sits above that for top-secret and especially sensitive national-security systems. So when the Pentagon says these deals reach IL6 and IL7, it is saying commercial AI vendors are being pulled deeper into the military’s most protected digital environments. ### Why is Anthropic missing? That is the part turning this into more than a contract story. Defense CTO Emil Michael said on May 1 that Anthropic is still considered a supply-chain risk by the department. He also said defense contractors must certify they are not using Anthropic’s Claude models in military work. The split appears to come from issues like domestic mass surveillance and autonomous weapons. ### Then what is the Mythos carveout? Michael said Anthropic’s Mythos model is being treated as “a separate national security moment.” His point was that Mythos has unusual cyber capabilities — especially around finding vulnerabilities and patching them — and that this is being handled across government, not just inside the Pentagon. The catch is that this does not look like it can separate one strategically useful model from a broader dispute with the company behind it. That last part is an inference, but it fits the public statements. ### Why sign eight deals instead of one big one? Basically, the Pentagon is trying to avoid vendor lock-in while moving fast. One of the public messages around the rollout is that the department wants a diverse AI stack rather than a single dominant supplier. That makes sense in military procurement — different models are better at different tasks, and dependence on one company becomes a strategic weakness if pricing, policy, or politics change. ### Is this only for classified missions? No — but classified access is the real escalation. The Pentagon’s broader GenAI.mil platform has already been used by more than 1.3 million Defense Department personnel, mostly for non-classified work like drafting, research, and analysis. The new agreements push beyond that safer sandbox and into the systems that support intelligence, warfighting, and core enterprise operations. ### What should companies take from this? The message is that capability alone is not enough. Model behavior, contract terms, safety restrictions, and supplier trust are now procurement issues, not side debates. If you sell into defense — or sell to someone who does — your AI stack can become a compliance problem overnight if a vendor falls out of favor. ### Bottom line The Pentagon is not slowing its AI adoption. It is widening the supplier list, pushing commercial models into classified systems, and making trust a hard gate. That is good news for the winners of this round. But it also means defense AI is becoming a two-front contest — one over model performance, and one over who the government believes it can safely depend on.