Army Burns Through Year's AI Tokens in One Month

On May 2026, the Army announced it would provide unlimited access to AI tokens for its Combat Capabilities Development Command (DEVCOM) and other units. The service chose Ask Sage, a multimodal generative AI platform that allows users to run multiple large language models: Alphabet's Gemini, Meta's Llama, and OpenAI's ChatGPT. Ask Sage is accredited for Controlled Unclassified Information and has been used by the Army's Chief Digital and AI Office (CDAO) for acquisitions, as well as for tasks like reclassifying personnel descriptions and aligning job duties and backgrounds.

To support this rollout, the Army purchased an enterprise pack subscription that provided 100,000,000 tokens annually. (A token is a unit of output—text or image—from a large language model; for Ask Sage, a single token represents about 3.7 characters.) Workers were initially allocated at least 200,000 tokens per month, and the system was configured to automatically grant additional tokens if an employee exceeded their monthly limit. The Army also encouraged usage through other means: workers who signed up but were not regularly using Ask Sage received emails urging them to consume more of their allocated tokens.

By mid-June 2026—less than one month after unlimited access was announced—the Army had burned through its entire year's supply of tokens. An internal email, viewed by WIRED, states: "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." According to an Army employee who spoke to WIRED on condition of anonymity, "the whole Army burned through the whole year of tokens for just one service." The email goes on to note that although the Army has renewed token usage "at its current levels," it remains unclear "if the Army CIO pool will be renewed after 1 Oct."

This pattern mirrors similar experiences at other large organizations. Meta, after encouraging employees to "tokenmaxx," quietly removed a leaderboard that tracked token usage and is now attempting to limit consumption. Adam Mosseri, head of Instagram at Meta, recently proposed capping token use per engineer. Uber experienced a similar dynamic: its engineers consumed a full year's worth of generative AI tokens in just four months. The Army employee noted that despite heavy organizational push, the AI tools have not proved particularly useful in practice. "When they have used the tools, they have found them to be unreliable," the article states, including instances where a model falsely claimed to have completed a task. The employee concluded: "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."

Meanwhile, the Pentagon continues expanding its AI ambitions. The Defense Department has cut staff at the Civilian Protection Center of Excellence, a unit responsible for preventing civilian casualties, and is instead developing an AI tool to speed up assessments that the Center's staff would normally perform. Neither the Army, the DOD, nor Ask Sage replied to requests for comment. The uncertainty over token renewal after October 1 leaves the future of Ask Sage access unresolved for Army workers.

The Army's token depletion reveals a fundamental mismatch between institutional enthusiasm for generative AI and the actual resource constraints of deploying it at scale. In May 2026, the Army announced unlimited token access to 3.5 million employees, framing AI adoption as a strategic priority. Yet within weeks, the annual budget—100,000,000 tokens—was exhausted. The article suggests this was partly driven by deliberate encouragement: workers received monthly allotments of at least 200,000 tokens with automatic top-ups, and inactive users were sent emails pushing them to consume more. This mirrors Meta's internal "tokenmaxx" culture, which the company has since quietly walked back. The parallel examples of Meta and Uber, both of which saw workers burn through annual AI token budgets in far shorter timespans, suggest the Army's experience is not isolated; once access becomes frictionless, consumption scales far beyond initial planning assumptions.

The broader context is one of Pentagon-wide organizational shift. Simultaneously with the token crisis, the Defense Department has cut staff at the Civilian Protection Center of Excellence—a unit focused on preventing civilian casualties—and is developing AI tools to automate assessments those staff members would normally perform. This suggests the Pentagon sees generative AI not as a supplementary tool but as a replacement for human roles. Yet an Army employee quoted in the article reports that the AI tools have been unreliable in practice, including false claims of task completion, and questioned whether an "unthinking application" would yield a trustworthy system. The uncertainty over token renewal after October 1 leaves the Army in a holding pattern, unable to commit to either a permanent cap or a budget increase.