The Pentagon added ChatGPT Mil and Grok for Government to its GenAI.mil platform on August 31 — quietly, with no press releases from OpenAI or xAI. The Department of War made the announcement itself. The platform now gives more than 1.7 million active military and civilian personnel access to three frontier AI models running side by side: Google Gemini, ChatGPT Mil, and Grok for Government. Three models, one secure platform, zero vendor lock-in — and one very conspicuous absence.
What GenAI.mil Is (and Why It Grew This Fast) #
GenAI.mil launched in December 2025 under Defense Secretary Pete Hegseth, initially running only Google Gemini for Government. The platform is not for classified data. Its purpose is to give the Pentagon’s 3 million military and civilian personnel a government-sanctioned place to use frontier AI on Controlled Unclassified Information (CUI) without routing sensitive data through consumer tools. It is the alternative to using ChatGPT.com and hoping for the best.
Adoption tells the real story: 80,000 users at launch, 1.5 million daily active users by mid-2026. The platform added ChatGPT Mil and Grok on August 31, pushing the active user base to over 1.7 million. Every model on the platform must carry Impact Level 5 (IL5) accreditation, a cloud security standard that most commercial AI platforms have not pursued.
What IL5 Actually Requires #
IL5 sits above FedRAMP High. It adds 10 additional security controls, requires physical network separation from non-government tenants (logical isolation is not sufficient), mandates continuous monitoring with defined remediation timelines, and prohibits using government data to train models. In practical terms: the AI you are talking to cannot learn from what you tell it, and the infrastructure it runs on cannot share space with commercial tenants.
Most AI providers have not bothered pursuing IL5 because the certification process is expensive, time-consuming, and requires infrastructure investment that does not make sense for a market that — until recently — was not large enough to justify it. The three providers now on GenAI.mil are the three that did the work. The full DoD IL5 requirements are documented on Microsoft Learn for those who want the technical specifics.
Three Models, One Platform, Deliberate Redundancy #
The Pentagon is explicit about its strategy: running multiple models simultaneously is a deliberate hedge against vendor lock-in. Gemini remains the original, general-purpose option. ChatGPT Mil brings the familiar interface most personnel already know — chat, files, projects, custom GPTs for repeatable workflows like filling out travel forms or drafting contracting responses. Grok for Government, from Elon Musk’s Starshield AI subsidiary, differentiates itself with deep-thinking inference, three reasoning modes (Auto, Fast, Expert), and persistent playbooks that capture institutional knowledge and make it reusable across teams.
Neither OpenAI nor xAI announced their own deployments. The Department of War made both announcements. That asymmetry is worth noting: these are significant enterprise contracts, and both vendors stayed quiet about them.
Claude Is Conspicuously Absent #
Anthropic’s Claude was the first frontier model cleared for classified military networks, approved in July 2025. Claude was ahead of everyone on this. Then, in January 2026, the Pentagon demanded that Anthropic waive Claude’s use restrictions to permit mass domestic surveillance and fully autonomous weapons systems. Anthropic CEO Dario Amodei said the company cannot in good conscience accede to those demands. Defense Secretary Hegseth disagreed. By February 27, 2026, the Pentagon had designated Anthropic a supply chain risk, and President Trump directed all federal agencies to stop using Anthropic’s technology.
Anthropic sued. Courts initially sided with the Pentagon. Then, on August 27 — just five days before this GenAI.mil expansion — US District Judge Rita Lin ruled the supply-chain designation unlawful, calling it First Amendment retaliation and arbitrary and capricious. The legal fight is not over, but Anthropic has now won the first significant court ruling. Claude still operates on classified networks under the original 2025 contract, but it is not on GenAI.mil. The model that got there first is now excluded from the unclassified platform.
The Enterprise Blueprint #
If you work on AI systems in any regulated industry — finance, healthcare, legal, insurance — GenAI.mil is the clearest preview of where enterprise AI procurement is heading. The requirements will look familiar: data isolation, no model training on user inputs, multi-vendor strategies to avoid single points of failure, and accreditation frameworks that take months to clear. The government is simply the most demanding early adopter, and what survives its requirements is what enterprises will eventually demand from their own providers. The IL5 moat is real. Providers who clear it gain a structural competitive advantage in the largest, most compliance-heavy market segment. The three models on GenAI.mil did not get there by being the smartest or cheapest. They got there by doing the compliance work that most AI companies are still deferring. Fortune has more detail on the full scope of the Pentagon’s AI ambitions.