In December 2025, the Pentagon gave 3 million people access to Google Gemini through GenAI.mil. By February 2026, 1.1 million had actually used it. ChatGPT and Grok joined the platform weeks later.
This is not a pilot program. This is not a study. This is over a million military and civilian personnel with frontier AI models on their desktops, backed by major contracts with each vendor.
All three military departments — Army, Navy, and Air Force — have announced adoption. Secretary Hegseth's AI Acceleration Strategy mandates that vendors deploy their latest models within 30 days of public release.
Now what?
The IL5 Ceiling
Here's what the press releases don't emphasize: GenAI.mil currently runs at Impact Level 5 — that's CUI (Controlled Unclassified Information), not classified. The Secret-level (IL6) transition is being actively pursued, but isn't operational yet.
So right now, the most common use case is probably the same thing everyone else uses ChatGPT for: drafting emails, summarizing reports, and writing slide decks.
That's not transformation. That's a very expensive autocomplete.
The real transformation happens when AI can process program data: cost reports, schedule data, risk registers, requirements traceability. When a program analyst can ask, "Show me every requirement that changed in the last 6 months and map the schedule impact."
That data lives on classified networks. And GenAI.mil's broad deployment is currently at IL5 — classified-network AI deployment is still emerging.
The Race to IL6
The Pentagon isn't standing still. Google achieved IL6 authorization for its Distributed Cloud infrastructure in 2025. The Army's Ask Sage platform was pushing toward IL6 deployment. Reuters reported in February 2026 that Pentagon CTO Emil Michael told tech executives the military is aiming to deploy frontier AI "across all classification levels."
But authorization and deployment are different things. Getting a cloud infrastructure certified at IL6 is step one. Making AI models available and useful on air-gapped classified networks — with all the data governance, access controls, and operational security that implies — is a much larger challenge.
The Data Infrastructure Gap
Even when IL6 AI arrives, the data infrastructure has to support it. Most defense program data lives in:
- Spreadsheets that nobody version-controls
- SharePoint sites organized by whoever set them up in 2014
- PDFs of PDFs of screenshots of slides
- Emails that contain critical decisions but aren't in any system of record
AI is only as useful as the data it can see. And right now, the data most defense programs produce is structured for human tolerance, not machine consumption.
The question isn't whether the DoD will adopt AI. That's done. The question is whether the data infrastructure can support it — or whether we'll spend 5 years teaching Gemini to write better PowerPoints while the actual program data remains trapped in formats AI can't touch. And while we figure out how to use AI defensively, adversaries are already using it offensively — deepfake voice attacks are coming to defense, and most commands don't have verification protocols ready.
What Comes Next
The 30-day deployment mandate from Secretary Hegseth's January 2026 memorandum is genuinely ambitious. It represents a sharp break from the days when it took Microsoft 18 months to make GPT-4 available in the Azure Government Top Secret cloud. But as Replicator showed us, ambitious mandates and actual delivery are different things.
But speed of model availability isn't the same as speed of useful deployment. The models are getting to the network faster. The question is whether the network has anything useful for them to do once they arrive.
For program analysts like me, the opportunity is enormous: imagine asking an AI to cross-reference your EVM data with your risk register, map dependencies across WBS elements, and flag the three most likely schedule drivers — in seconds instead of days.
We're not there. But we're closer than most people think. And the analysts who learn to structure their data for AI consumption now will be the ones who benefit first when IL6 GenAI finally arrives.
If you're a program analyst sitting on messy SharePoint data and PDF-of-PDF deliverables, start structuring that data now. The classified AI is coming. The analysts who are ready for it will leapfrog the ones still waiting for instructions.
Quick Answers
What is the core argument of GenAI.mil?
The core argument is that access to commercial AI at IL5 is not the same as deploying useful AI where the most sensitive military work happens, which still requires IL6-capable infrastructure and data readiness.
Why is data infrastructure as important as model access?
Because even if classified AI tools arrive, they cannot improve programs if the underlying data remains fragmented across spreadsheets, PDFs, email chains, and poorly structured SharePoint systems.
Sources: DefenseScoop (Feb 2, 2026), Breaking Defense (Feb 10, 2026), Secretary Hegseth AI Strategy Memo (Jan 9, 2026), Reuters (Feb 11, 2026), Google Cloud IL6 Authorization (Jan 2026)