Pentagon awards AI deals, Anthropic excluded
- On May 1, the Pentagon named seven AI vendors for classified-network work: OpenAI, Google, Microsoft, AWS, Nvidia, SpaceX, and Reflection. - The telling detail is who is missing. Anthropic had been approved for classified military use, but the Pentagon labeled it a supply-chain risk. - That turns this into more than a vendor shuffle — it is a test of whether defense access now depends on looser AI-use terms.
The Pentagon just made a very clear choice about who gets to help build military AI inside classified systems. OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, SpaceX, and Reflection are in. Anthropic is out. That matters because this is not just another cloud procurement story — it is about which model makers are willing to let the government use their systems under broad defense terms, and which ones are not. (techxplore.com) ### What actually got awarded? On May 1, the Department of War posted a release for “Classified Networks AI Agreements,” and outside coverage tied that announcement to seven named companies: SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, and AWS. The work is aimed at putting commercial AI tools onto the Pentagon’s most sensitive cl(techxplore.com)areness, and other defense tasks. (war.gov) ### Why is Anthropic the real story? Because Anthropic was not some random outsider. The Pentagon had already been using Claude in classified military operations, and reporting says Claude had been the only model authorized there for a time. Then the relationship broke down after Anthropic objected to uses involving mass domestic surveillance or direct control of lethal autonomous (war.gov)as a supply-chain risk, which effectively pushed it and its contractors out. (techxplore.com) ### What does “supply-chain risk” mean here? Basically, it is a procurement and security label that says the government should not depend on that vendor. In normal defense language, that kind of designation is associated with cybersecurity exposure, foreign influence, or reliability concerns. Using it against a major U.S. AI lab over policy terms is why this has turned into a much bigger fight than a lost contract. (msn.com) ### Why did the Pentagon want more vendors anyway? Part of the answer is simple — vendor concentration. Emil Michael, the Pentagon’s de facto chief technology officer, said it was irresponsible to rely on any one partner. The de(msn.com)ths. This new vendor lineup looks like the classified extension of that same strategy. (techxplore.com) ### Why are Nvidia and Reflection notable? Because they appear to give the Pentagon something different from the big closed-model vendors. Reporting says the deals include both closed and open-source models, with Nvidia and Reflection tied to the open side. That matters because open models can be run without constant vendor access or recur(techxplore.com)ability, and continuity matter more than glossy consumer features. (techxplore.com) ### Is this about policy as much as technology? Yes — maybe more. The Defense Department still has responsible-AI language on the books about lawful and ethical use, but the current push is plainly toward an “AI-first” military posture and faster operational deployment. So the live question is not whether the Pentagon wants AI. It clearly (techxplore.com)ve into defense workflows. (media.defense.gov) ### What happens next? Anthropic is already fighting the government’s actions in court, so this will likely move on two tracks at once — procurement and litigation. If the Pentagon keeps broadening classified AI access while excluding the one major lab that tried to hold a harder line on weapons and surveillance uses, other vendors will read the signal. Cooperation gets you inside. Guardrails may get treated as a risk. (techxplore.com) ### Bottom line? The Pentagon did not just pick vendors. It picked a model for how military AI gets bought and governed. And right now, the model favors firms willing to go where defense demand is heading.
