The US Army Ran Out of AI Credits. Here Is What That Tells Us.
The Army promised its workers unlimited AI access. Six weeks later, the credits were gone. Now soldiers and civilian staff face new limits, and one worker told Wired the tools were unreliable anyway.
Key points
- The US Army's annual allotment of 100 million AI tokens ran out by mid-June 2026, roughly six weeks after the Army's chief information officer announced unlimited access.
- Ask Sage, the Army's AI platform, lets staff run several large language models including Alphabet's Gemini and OpenAI's ChatGPT.
- Each employee received at least 200,000 tokens per month, with automatic top-ups if they ran out.
- Whether the Army's token pool will be renewed after 1 October 2026 is uncertain.
- Meta and Uber have hit similar walls after pushing heavy AI adoption.
The US Army spent the first half of 2026 telling its workers to use AI as much as possible. In May, the Army's chief information officer announced unlimited access. By mid-June, the entire year's supply of credits had gone.
An internal email, first reported by Wired, confirmed the shortfall and put limits back in place. "The Army CIO pool was exhausted of tokens," it read. "Apparently the whole Army burned through the whole year of tokens for just one service," an Army employee told Wired.
Tokens are the basic unit of work an AI model performs: every word it reads or writes consumes a few of them. One token covers roughly 3.7 characters on Ask Sage. The Army had bought 100 million as part of an annual subscription. Gone in under two months.
What does this mean for Army staff and their work?
Limits are back for now, but there's no guarantee the pool refills when the government's financial year ends on 1 October. Staff who've woven these tools into everyday tasks don't know whether they'll still have access.
One Army employee, speaking to Wired anonymously because they weren't authorised to talk to the press, said they hadn't found the tools reliable. One model told them it had finished a task it hadn't touched. "An unthinking application and use is not going to result in an effective, efficient, trustworthy rollout," they said.
Ask Sage is a multimodal generative AI platform connecting to large language models including Alphabet's Gemini, Meta's Llama, and OpenAI's ChatGPT. It's approved for controlled but unclassified government information and has been used for tasks like rewriting job descriptions. We covered Ask Sage for the first time on 21 July 2026.
The Army, the Department of Defense, and Ask Sage didn't reply to requests for comment.
Should you worry about AI credit limits spreading?
The Army isn't alone. After encouraging engineers to maximise token use, Meta quietly removed the leaderboard tracking who used the most; Instagram chief Adam Mosseri has since floated per-engineer caps. Uber reportedly burned through a full year of AI credits in four months.
The pattern holds across big organisations: push hard adoption, staff respond, bills arrive faster than budgets planned. Whether the Army's experience produces a more careful rollout or simply a larger budget request is the real question to watch. My read: the reliability problem that anonymous employee described matters more than the cost overrun. Credits can be bought. Trust in tools that hallucinate completed tasks is harder to rebuild.
