The Pentagon is quietly asking the biggest artificial intelligence players to bring their most advanced systems onto some of the United States’ most sensitive computer networks. The push reflects a rapid shift from pilot projects to classified deployment, as defense leaders race to plug commercial AI into war planning, intelligence work and cyber operations. At the same time, it raises hard questions about how far the military should lean on private models that were built for consumer apps, not nuclear command chains.
Behind closed doors, senior defense officials have been pressing companies that build large language models and other AI tools to harden their software for use on secret and top secret systems. They are targeting firms that already work with the Pentagon on unclassified projects, and asking them to scale up into secure environments that most Silicon Valley engineers never see. Observers inside the department describe this as a turning point, when the United States starts to treat commercial AI not as an experiment on the edge of the force, but as core infrastructure for national defense.
The classified AI push moves from talk to action
People familiar with the discussions say the Pentagon has urged top AI developers to deploy their systems inside secure facilities that handle classified data, including intelligence feeds and operational plans. Those efforts focus on large language models that can summarize reports, draft plans and sift through huge data sets, as well as other tools that can scan satellite images or detect cyber threats. The pressure on companies to expand into these protected environments has been described as a coordinated effort by senior defense leaders, rather than a series of isolated contracts, and several sources who requested anonymity have outlined that shift in classified expansion.
The campaign builds on earlier outreach to major model makers such as OpenAI and Anthropi, whose systems already underpin many commercial and government pilots. According to people briefed on the conversations, the Pentagon has framed the move into classified networks as a natural next step for companies that want to stay at the center of defense AI strategy. Reporting that cites By David Jeans and Deepa Seetharaman Feb, and references to the Pentagon’s Chief Technology Officer Emil, describes how senior officials have personally pushed OpenAI and Anthropi to bring their tools into secure environments, a sign of how central these firms have become to the military’s AI plans and how much weight leaders now place on these discussions.
From pilot contracts to deep integration with OpenAI, Anthropic, Google and xAI
The move into classified systems did not appear out of thin air; it follows a series of large unclassified deals that pulled commercial AI into the heart of the Pentagon’s technology stack. In one major step, the department awarded multiple companies contracts worth a combined $200,000,000 for AI tools, a package that went to Four tech firms: Anthropic, Google, OpenAI and xAI. Those awards, overseen by the Chief Digital and AI Officer Dr., signaled that the military saw these four firms as core partners for everything from planning tools to decision support, and they set the stage for a deeper push into secure networks by giving each company a foothold inside military systems.
That same summer, The Pentagon announced Monday that it had chosen Google, xAI, Anthropic and OpenAI to help expand the military’s use of advanced AI. The selection, also tied to the Chief Digital and AI Office, formalized the role of those companies as strategic partners rather than short term vendors, and it marked the point when commercial models began to look like standard equipment for U.S. forces. By bringing Google, Anthropic and the others into a structured program, the department created a pipeline for their tools to move from test ranges into real missions, and it is that same group of firms that is now being asked to adapt their systems for classified deployment.
GenAI.mil, Gemini for Government and the road to secret networks
As the Pentagon built those partnerships, it also stood up a dedicated environment for AI experimentation and use, a site known as Genai.mill. Earlier in the current cycle, Defense Secretary Pete Hegseth said that Google Cloud’s Gemini for Government would be the first AI capability launched on that platform, and described the move in a video posted on social media Tuesday. By putting Gemini for Government on Genai.mill, the department created a controlled space for sensitive but unclassified work, and a stepping stone toward the even tighter security that secret and top secret networks demand, with Hegseth presenting the move as a major boost for AI adoption.
The Genai.mill effort has since expanded beyond Google, as the Pentagon invited more model providers into the environment. In a separate announcement, US Secretary of Defense Pete Hegseth said the department was excited to bring the next frontier AI model company to Genai.mill, naming Grock from XAI as a system that would go live on the site later in the year. That move pulled XAI’s technology into the same pipeline as Gemini for Government, and it showed how Genai.mill is becoming a staging ground where tools from Google Cloud, Grock and other firms can be hardened before they move into more sensitive spaces, with Hegseth casting Grock and XAI as central players in the Pentagon’s AI partnerships.
Elon Musk’s Grok and the leap to fully classified systems
One of the most striking developments is the Pentagon’s embrace of AI tied to Elon Musk, who has long been a vocal critic of some government policies even as his companies supply rockets, satellites and cars. A short video clip describes how the Pentagon is putting Elon Musk’s AI inside some of its most secret military systems, framing the move as either the future of defense or a huge risk to humanity. The clip focuses on a model called Grok, which is associated with Musk’s ventures and is now being adapted for defense use, raising questions about how far the department should rely on a tool that was born in the tech world and carries its creator’s strong public persona into military planning.
The same Grok branding appears in the Genai.mill rollout, where US Secretary of Defense Pete Hegseth called Grock from XAI the next frontier AI model company to join the platform. While the spelling in that announcement uses Grock rather than Grok, the association with XAI and with Musk’s broader ecosystem is clear, and it shows how quickly tools linked to high profile tech figures can move from social media chatter to defense infrastructure. By bringing Grock from XAI into Genai.mill, and by testing Elon Musk’s AI on secret systems, the Pentagon is tying its classified AI future to a set of commercial brands that were built for public markets, not secure bunkers, and that choice will shape debates over control, reliability and long term risk.
Security, control and the quiet debate inside the Pentagon
Inside the building, the rush to bring OpenAI, Anthropi, Google, xAI and others onto classified systems has sparked a quiet argument over security and control. Some officials argue that the United States cannot afford to leave its best AI power sitting on commercial clouds while rivals race to arm their own militaries with similar tools. Others worry that plugging large language models into secret networks could expose sensitive data if the systems are not properly isolated, or if model providers retain too much access to logs and training material. Reporting that describes Exclusive coverage of the Pentagon’s push, including references to FILE and PHOTO of The Pentagon logo, captures how seriously leaders now treat both the promise and the danger of putting commercial AI in charge of classified work.
