The Defense Department’s push to embed generative AI in everyday workflows hit a stark reality check during Operation Epic Fury, a 38‑day campaign in Iran. The Army’s Ask Sage subscription—an "enterprise pack" that grants 100 million tokens a year—was dwarfed by a consumption rate of about 20 billion tokens each day, according to documents reviewed by WIRED and Breaking Defense.
A token represents roughly 3.7 characters of output, whether text or image, so the daily drain translates to an output volume that would fill thousands of books in a single day. The Army’s own personnel say the tools have been little more than a novelty, with some models falsely claiming to have completed tasks they never performed.
Department officials declined to comment, and Ask Sage did not respond to requests for clarification. It remains unclear whether the tokens burned by field operators draw from the same pool allocated to civilian DOD employees working on classified or secret projects.
Despite the runaway usage, the Pentagon has not slowed its AI ambitions. Earlier this week, the Intercept reported a reallocation of resources from the Civilian Protection Center of Excellence—an office tasked with reducing civilian casualties—to the development of a new AI system designed to automate the very assessments the Center once performed manually.
The Army’s experience mirrors a broader trend of over‑enthusiastic token consumption across tech firms. Meta, after encouraging engineers to "tokenmaxx," quietly removed its internal leaderboard tracking usage and is now considering caps on token allocations per engineer. Uber, too, reportedly burned through a year's worth of generative‑AI tokens in just four months, according to Fortune.
One Army employee, speaking on condition of anonymity, described the generative tools as “unreliable” and “not particularly useful” for mission‑critical work. The same source warned that an “unthinking application” of AI could undermine efficiency, trust, and ultimately, effectiveness within the sprawling federal bureaucracy.
As the DoD continues to experiment with AI, the stark token depletion in Iran may force a recalibration of expectations. The episode underscores that while generative models promise rapid content creation, the practical limits of token‑based licensing and the need for robust oversight remain significant hurdles for military adoption.
Este artigo foi escrito com a assistência de IA.
News Factory APP - notícias agênticas para impulsionar seu SEO e AEO.