The Army Said 'Unlimited AI Tokens.' Narrator: It Was Not.
What happened
Ask Sage, the Army's enterprise LLM platform, gave soldiers what the CIO called 'unlimited tokens' starting May 2026. By mid-June, the entire year's 100-million-token enterprise pack was gone, and DEVCOM staff got an email telling them to ration usage. One Army employee told WIRED bluntly: the whole Army burned through a year's tokens for one service branch.
Why this matters
This isn't a rounding error — it's a live case study in what happens when a massive bureaucracy tells hundreds of thousands of employees to 'lean into AI' without metering reality-checks. It also reveals how opaque token economics are even to the people paying for them: nobody seems to know if classified-use tokens draw from the same pool as HR paperwork.
The slightly cynical read
The DOD spent weeks bragging that half its 3.5 million employees were 'using AI,' which sounds great until you realize 'using AI' apparently means burning a year's compute budget reclassifying personnel job titles. Meanwhile the Pentagon is simultaneously gutting the team that assesses civilian casualties in favor of an AI tool — priorities, clearly, are elsewhere.
What to watch next
Whether the Army CIO renews the token pool after October 1st, or whether soldiers go back to email and PowerPoint. Also worth watching: whether Congress asks why nobody can say if classified AI use shares a token pool with routine paperwork.
