The US Department of War on Friday signed AI agreements with seven of America’s largest tech and infrastructure companies to deploy frontier models on classified networks.
The contracts cover SpaceX, OpenAI, Google, NVIDIA, Reflection AI, Microsoft and Amazon Web Services. They authorize the firms’ AI to operate inside Impact Level 6 and Impact Level 7 environments for any lawful operational use.
Inside the Department of War’s AI Agreements
The Department’s Chief Technology Officer announced the package on May 1 and framed it as the latest step in building what officials call an “AI-first” War Department. IL6 and IL7 designations cover secret and top-secret workloads, so the models will sit alongside sensitive intelligence and operational data.
“This is just the latest initiative in our mandate to create an AI-FIRST WAR DEPARTMENT,” the official account for the Office of the Under Secretary of War for Research and Engineering stated.
Officials said the spread of vendors is intentional. By contracting with multiple US providers, the Department aims to avoid vendor lock-in and keep options open across closed and open-source models.
NVIDIA’s portion includes its open-source Nemotron family, while Reflection AI, an Nvidia-backed startup founded by former Google DeepMind researchers, will supply additional open-weight systems.
Google brings its Gemini family for any lawful government purpose, and SpaceX is expected to contribute infrastructure tied to xAI’s Grok models.
Microsoft and AWS keep their roles as cloud and infrastructure backbone for the rollout.
Internal adoption is already heavy. The Department’s GenAI.mil platform has crossed 1.3 million users and tens of millions of prompts within five months of launch, according to the May 1 release.
Anthropic Sits Out After Guardrail Standoff
The roster does not include Anthropic. Defense Secretary Pete Hegseth labeled the company a supply-chain risk in February after Anthropic refused to remove restrictions on autonomous lethal weapons and mass domestic surveillance.
“We will not let ANY company dictate the terms regarding how we make operational decisions,” Defense Department spokesman Sean Parnell articulated.
A federal judge later blocked enforcement of the ban, and the legal fight continues.
OpenAI took a narrower path than rivals. The company said its War Department deal preserves three commitments:
- Its models cannot be used for mass domestic surveillance,
- Cannot direct autonomous weapons, and
- Will keep their safety guardrails in place.
Other firms accepted broader “any lawful purpose” language without those public carve-outs.
Open-Source Push Sets the Tone for What Comes Next
The deals fold into the Department’s AI Acceleration Strategy, published earlier in 2026, which calls for modular open-source architectures across warfighting, intelligence and enterprise functions.
Officials said the strategy favors domestic vendors, transparent open-weight options, and rapid prototyping over closed-model dependence.
The next watch points will be which models clear IL6 deployment first and whether OpenAI’s published guardrails hold up once classified workflows scale.
