Technology news and discussion on TalkFuse.
Unlimited AI tokens aren’t unlimited after all as US Army burns through supply
The US Army has had to reimpose limits on staff use of generative AI tools after burning through an entire year's supply of tokens in about a month, despite having announced "unlimited" access in May 2026. Members of the Army's Combat Capabilities Development Command were told by email that the CIO's token pool, offered as part of a subscription to the Ask Sage platform, was exhausted by mid-June and that usage would continue at current levels but might not be renewed after 1 October. The episode highlights the gap between the Pentagon's enthusiastic push for AI adoption and the practical costs and limits of running it at scale, particularly given uncertainty over future funding.
Ask Sage is a multimodal AI platform accredited for handling Controlled Unclassified Information, giving Army staff access to models including Google's Gemini, Meta's Llama and OpenAI's ChatGPT for tasks such as reclassifying personnel job descriptions. The Army held an annual allowance of 100 million tokens, with employees given at least 200,000 tokens a month and encouraged to use more if under-utilising their allocation; by comparison, the wider Defense Department reportedly used around 20 billion tokens a day during a 38-day military operation in Iran. The Army and Ask Sage did not respond to requests for comment, and an anonymous Army employee told WIRED they had found the tools unreliable. Similar overuse has hit other organisations, with Meta curbing its "tokenmaxx" culture and Uber reportedly exhausting a year's token budget in just four months.
In short
- Army exhausted a year's AI token supply within about a month
- "Unlimited" access announced in May 2026 quietly reversed by June
- Meta and Uber have faced similar AI token overuse issues
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The rapid token consumption demonstrates that AI tools are filling genuine operational needs across Army staff, indicating successful adoption of a Pentagon priority. The exhaustion reflects demand outpacing initial budget estimates—a common challenge when introducing transformative technologies. Rather than revealing a flawed initiative, this illustrates that realistic funding must match the actual organisational appetite for AI capabilities, which appears substantial and productive.
The case against
Announcing 'unlimited' access whilst working from a finite annual budget represents either dishonesty or gross miscalculation in project planning. Encouraging staff to use more tokens without communicating actual constraints wasted resources and undermined credibility. The anonymous report of unreliable tools raises questions about whether rapid consumption reflected genuine value creation or simply usage for its own sake, suggesting the initiative needs proper evaluation before expansion or renewal.
Explained simply
Generative AI tools produce text and other content in response to prompts. They process those prompts using tokens, small units of text that are counted towards a service’s usage allowance; the amount available can depend on the provider and subscription.
The US Army’s Combat Capabilities Development Command is one of the Army’s research and development organisations. Its staff use Ask Sage, a platform that provides access to AI models including Gemini, Llama and ChatGPT, for work such as revising job descriptions; the platform is approved for some information that is not classified but still requires protection.
The episode illustrates the practical challenge of expanding AI use across a large organisation: demand can exceed the allowance, and continued access depends on budgets and subscriptions. It also raises questions about how reliable and useful these tools are for staff, and how organisations should manage their use.