The US Army exhausted a 100,000,000-token annual subscription to Ask Sage within weeks of offering staff unlimited access, forcing the Army CIO to reinstate usage caps and warn employees that the pool may not be replenished after October 1. An email sent to the Army's Combat Capabilities Development Command, known as DEVCOM, informed workers that although unlimited tokens were announced in May 2026, the pool ran dry by mid-June. The disclosure comes barely a month after the Department of Defense said nearly half of its 3.5 million employees were using AI at work.
Ask Sage is a multimodal generative AI platform that lets Army users route prompts to models including Alphabet's Gemini, Meta's Llama, and OpenAI's ChatGPT. It is accredited for Controlled Unclassified Information and is also used by the DOD's Chief Digital and AI Office for acquisitions work. A single Ask Sage token corresponds to roughly 3.7 characters of model output.
“Although the Army CIO announced in May 2026 that they were offering unlimited tokens, by mid-June the Army CIO pool was exhausted of tokens and had to re-establish limits”— Army CIO office, internal email to DEVCOM staff
Individual Army employees were allotted at least 200,000 tokens per month, with automatic top-ups when they ran through the initial pool. Workers who signed up but did not use the tool were prodded by email to consume more of their allocation. That mandate-to-use, layered on top of a fixed annual ceiling, is a plausible explanation for how a 100M-token pack designed to last twelve months lasted closer to six weeks.
Key facts
- 01The Army's 100M-token annual Ask Sage subscription was exhausted within weeks of the CIO announcing unlimited access in May 2026.
- 02DEVCOM employees were originally allotted 200,000 tokens per month, with automatic top-ups if they ran out.
- 03The DOD burned through roughly 20 billion tokens per day during the 38-day Operation Epic Fury in Iran.
- 04Nearly half of the Department of Defense's 3.5 million employees are now using AI at work, per a June 2026 disclosure.
- 05Renewal of the Army CIO token pool after October 1 is uncertain, according to the internal email.
The Army's burn rate is small compared with the DOD's operational usage. According to reporting on Operation Epic Fury, the 38-day Iran operation, the Defense Department consumed roughly 20 billion tokens per day. It is not clear from public disclosures whether that operational pool draws from the same budget as the Army CIO's enterprise pack, or whether classified workflows sit on a separate contract entirely.
The exhaustion has not slowed the Pentagon's AI push. The DOD has continued to lean on generative tools even while cutting staff at the Civilian Protection Center of Excellence, whose analysts previously worked to prevent civilian casualties in conflict zones. The department is building an AI system to accelerate the assessments that team used to produce manually.
The Army is not alone in mispricing generative AI consumption inside a large organization. Meta, after encouraging engineers to "tokenmaxx," quietly retired its internal leaderboard tracking token usage and is now trying to curb consumption. Adam Mosseri, head of Instagram, has floated capping tokens per engineer. Uber engineers burned through a year of generative AI tokens in four months, according to Fortune.
Ask Sage's documented Army use cases include reclassifying personnel descriptions - defining and aligning job duties, experience, and backgrounds. That is the sort of paperwork-heavy task LLMs are well suited to, but it is also the sort of task where a single retry, a longer output, or a chatty system prompt can multiply token draw by an order of magnitude with no visible change to the user.
The DEVCOM staffer who described the shortfall to Wired said the tools have not been consistently useful, and that in one case a model claimed to have completed a task it had not. Reliability failures at that level are not a rounding error inside a defense workflow; they set the ceiling on what a token budget is actually buying.
“I think there are definitely several aspects of the bureaucracy of the US federal government that these tools might be helpful with. But an unthinking application and use is not going to result in an effective, efficient, and trustworthy rollout.”— Army employee, DEVCOM staffer speaking anonymously
The Army, the DOD, and Ask Sage did not respond to requests for comment. The email to DEVCOM staff said usage would be renewed at current levels but did not commit to funding beyond October 1, the start of the federal fiscal year, when a fresh budget line would be needed.
The Army episode is a preview of the procurement problem every large AI buyer is about to face. Enterprise token packs priced on last year's usage assumptions do not survive an executive mandate to "use more AI," and the vendors selling annual pools have a structural incentive to underestimate consumption. Expect metered pricing, hard per-seat caps, and workload-tier contracts to replace flat annual packs across federal AI procurement over the next twelve months - because Meta, Uber, and now the US Army have all just proven that unlimited access, offered to a workforce told to use it, is not a budget line any CIO can defend.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
