Army faces AI token limits after early surge in workplace use
An Army email said a CIO token pool was exhausted by mid-June after workers were urged to use the Ask Sage generative AI platform.
By James Whitfield · Staff Writer
3 min read
The Army has had to rein in access to generative AI tokens after a workplace rollout consumed a token pool far faster than expected, according to emails reviewed by WIRED. The episode shows how quickly everyday AI use can collide with usage caps and computing costs, even inside a Defense Department program promoted as broad access.
Members of the Army’s Combat Capabilities Development Command received an email saying the Army chief information officer’s token pool had been depleted by mid-June, WIRED reported. The message said the Army CIO had announced in May 2026 that workers would get unlimited tokens, but limits had to be restored after the pool ran out.
The email said the Army chose to renew token use at current levels, according to WIRED. It also said it was unclear whether the CIO pool would be renewed after Oct. 1.
Ask Sage use spread across the Army
The Army uses Ask Sage, a generative AI platform that gives users access to large language models including Google parent Alphabet’s Gemini, Meta’s Llama and OpenAI’s ChatGPT, according to WIRED and the Army. An Army employee who spoke anonymously to WIRED because they were not authorized to talk to reporters said the service appeared to have consumed a full year’s token supply.
Ask Sage says it serves defense customers, and the Army has described it as the basis for an enterprise large-language-model workspace. The Army has said the platform is accredited for controlled unclassified information and has been used for tasks including work on personnel descriptions.
According to WIRED, Army employees received at least 200,000 tokens a month, based on emails the outlet reviewed. Workers who used up that amount were automatically given more, while employees who had registered for Ask Sage but used it infrequently received messages encouraging more use.
The annual Ask Sage enterprise subscription gave the Army access to 100,000,000 tokens, WIRED reported. Tokens are units used to measure AI model input or output; for Ask Sage, one token equals about 3.7 characters, according to documents reviewed by WIRED.
Defense AI demand keeps rising
The Defense Department has pushed wider AI use across its workforce. A little more than a month before the DEVCOM email, the department said nearly half of its 3.5 million employees were using AI at work, according to WIRED.
Demand for AI tools has also appeared in military operations. Breaking Defense reported that the Defense Department used about 20 billion tokens a day during the 38-day Operation Epic Fury in Iran.
WIRED reported that it remains unclear whether tokens used by ordinary Defense Department employees come from the same pool as tokens used for tools handling classified or secret information. The Army, the Defense Department and Ask Sage did not respond to WIRED’s requests for comment.
The Intercept reported this week that the Pentagon has kept developing AI tools while cutting staff at the Civilian Protection Center of Excellence, whose work involved preventing civilian casualties in conflict zones. According to The Intercept, the Defense Department is building an AI system to speed assessments normally performed by that center’s staff.
Other large organizations have also pulled back after heavy token consumption. The New York Times reported that Meta removed a leaderboard that tracked employee token use after previously encouraging high usage, and TechCrunch reported that Instagram head Adam Mosseri raised the possibility of per-engineer token caps. Fortune reported that Uber engineers used a year’s worth of generative AI tokens in four months.
The Army employee who spoke to WIRED said the tools had not been very useful for their work and had sometimes been unreliable, including one model that claimed to have finished a task it had not completed. The employee said some federal bureaucracy may benefit from AI, but careless deployment would not produce a trustworthy rollout.
This story draws on original reporting from Ars Technica.