The US Department of War has formalised agreements with seven leading AI companies, including OpenAI, Google, NVIDIA, Microsoft, Amazon Web Services, SpaceX, and Reflection, to deploy frontier AI on its highest-classification military networks, a move that accelerates the transformation of the US armed forces into an AI-first fighting force.
The announcement dropped on April 30, and the breadth of it is striking. Seven of the most consequential companies in artificial intelligence, covering the full stack from cloud infrastructure to frontier models to specialised hardware, have all committed to deploying their capabilities inside the War Department's Impact Level 6 and Impact Level 7 network environments. IL6 and IL7 are the classification tiers that handle secret and top-secret military data respectively. These are not administrative networks for scheduling and email. They are the environments where mission planning, intelligence analysis, targeting decisions, and operational coordination actually happen.
The path to this moment was not smooth. As recently as February 2026, Reuters reported that the Pentagon's push to get frontier AI onto classified networks was running into significant friction. Pentagon CTO Emil Michael told tech executives at a White House event that the military wanted AI models available across all classification levels, including with reduced guardrails compared to what commercial users receive. Several companies pushed back. Anthropic's negotiations were described as notably contentious, with the company's leadership publicly resisting any application of their technology to autonomous weapon targeting or domestic surveillance. Anthropic is conspicuously absent from the seven companies named in April 30's announcement.
OpenAI formalised its classified network deal in late February. Sam Altman posted that evening that they had reached agreement to deploy their models on the Department of War's classified network. The company published a detailed rationale on its website, outlining three core red lines it would not cross regardless of instructions, and noting that forward-deployed OpenAI engineers with security clearances would remain involved in every classified deployment alongside cleared safety researchers. The framing was careful and deliberate, designed to demonstrate that OpenAI was entering this territory with more structural oversight than any prior classified AI agreement, including Anthropic's earlier arrangement. Google finalised its own classified deal on April 28, two days before the broader seven-company announcement, covering use of its Gemini models for any lawful governmental purpose across a wide range of classified systems.
The technology being deployed across these environments will support three areas the War Department identifies as central to its AI Acceleration Strategy: warfighting, intelligence, and enterprise operations. In practical terms, that means AI systems that can synthesise intelligence from multiple classified sources faster than human analysts, support decision-making in complex operational environments where time and information density are both critical, and streamline the logistics and operational planning functions that underpin military effectiveness. The phrase "decision superiority" appears in the official release. It is the military's way of describing the advantage that comes from processing and acting on information faster than an adversary.
NVIDIA's inclusion alongside the model providers is telling. It positions GPU infrastructure as equally foundational to the military's AI strategy as the software layer. Running frontier models at classification levels that require air-gapped or highly secured environments demands specialised hardware configurations that NVIDIA is uniquely positioned to provide. SpaceX brings satellite communications infrastructure through Starlink, relevant to deployments in contested or remote environments. Reflection, less publicly prominent than the others, is a startup developing frontier models and its inclusion signals how comprehensively the Pentagon is diversifying its AI supplier base rather than relying on any single large provider.
The ethical dimensions here are real and deserve more than a footnote. Deploying large language models in environments where outputs could influence targeting decisions or intelligence assessments introduces failure modes that are genuinely different from commercial applications. A hallucinated output in a consumer product is an annoyance. The same failure mode in a classified operational environment carries different consequences. This is precisely why OpenAI's insistence on maintaining cleared engineers in the loop for classified deployments, and its published red lines around weapons of mass destruction and AI acting without human oversight in consequential decisions, represents the kind of structural safeguard the technology requires at this level of deployment.
What the April 30 announcement confirms is that the US has decisively accelerated its position in a military AI race that China is running in parallel. The People's Liberation Army has publicly committed to AI integration across its command and control systems. The Pentagon's move to bring the full depth of American frontier AI onto its most classified networks is a direct response to that trajectory. The companies involved have concluded, after months of negotiation, that the mission requirements and their own safety frameworks are compatible enough to proceed. The world's most powerful military is now an active customer of the world's most advanced AI, on its most sensitive networks. That is a significant threshold to have crossed.
Also read: Nebius acquires Eigen AI for $643 million as the inference bottleneck becomes the new GPU war • Retail traders are handing their portfolios to AI agents and the early results are as mixed as you would expect • Synaps raised €3 million to take on AutoCAD with an AI-native architectural platform and the bet is bigger than it looks
