On May 1st, the Pentagon made a historic announcement: eight tech companies are now cleared to deploy their AI models on the most classified networks in the US military. OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, SpaceX, the startup Reflection, and Oracle — they’re all in. One name is conspicuously absent: Anthropic.
What was actually agreed?
The deals allow these companies to run their AI systems on so-called Impact Level 6 and Impact Level 7 networks. IL6 covers secret-level information, IL7 handles the most sensitive intelligence data. Pentagon CTO Emil Michael called it a “transformation toward an AI-first fighting force.”
The Pentagon’s condition: all models must be usable for “any lawful purpose” — with no restrictions imposed by the vendors.
Why Anthropic isn’t on the list
And that’s exactly the sticking point. Anthropic set conditions for the Pentagon: Claude should not be used for fully autonomous weapons or mass surveillance. The company didn’t budge on these red lines — and Defense Secretary Pete Hegseth classified Anthropic as a “supply chain risk” back in February. That’s a label normally reserved for foreign adversaries.
SpaceX as a surprise entry
Elon Musk’s SpaceX making the list is noteworthy. SpaceX is primarily a rocket company, not an AI lab. But with the planned integration of xAI — Musk’s AI venture — into SpaceX, the deal suddenly makes sense. Particularly interesting timing, given the ongoing Musk vs. Altman trial happening in parallel.
The bigger picture
The situation is deeply paradoxical. The company that invests the most in AI safety gets excluded from the Pentagon — precisely because it insists on safety guardrails. Anthropic’s Mythos model, which has discovered thousands of zero-day vulnerabilities, is already used by the NSA. But the Pentagon still wants nothing to do with the company.
For Anthropic, this is an expensive principled stand. For the AI industry, it’s a fundamental question: how much say should the builders have when their models go to war?
Sources: