The US Army has been forced to reimpose limits on staff use of generative AI after burning through its allotted token pool far faster than expected, according to internal emails seen by WIRED. Members of the Army's Combat Capabilities Development Command were told the CIO's promised "unlimited" tokens, announced in May 2026, had already been exhausted by mid-June, prompting a return to capped usage. The episode highlights the gap between the Pentagon's public enthusiasm for AI adoption and the practical costs of running it at scale.

The Army uses the platform Ask Sage, which lets staff access models including Google's Gemini, Meta's Llama and OpenAI's ChatGPT, and held an annual "enterprise pack" subscription covering 100 million tokens. Employees had been encouraged to use at least 200,000 tokens a month, with more allocated automatically and reminder emails sent to those who weren't using their quota. It remains unclear whether the Army's pool will be renewed after 1 October. The Army is not alone in this: Meta has quietly scrapped an internal leaderboard that had encouraged staff to "tokenmax", and Uber reportedly saw engineers exhaust a year's worth of tokens in just four months.