Overview
- Internal Army emails verified by Wired show leaders reimposed token caps after the service used up an annual 100‑million token allotment within weeks.
- The consumption happened on Ask Sage, the Army’s secure gateway to commercial models such as ChatGPT, Gemini and Llama, which troops used for tasks like rewriting job descriptions and processing procurement.
- The Army had given roughly 200,000 tokens per person per month and set automatic top-ups for heavy users, a policy that accelerated depletion of the enterprise pack.
- Reported operational spikes in use — including a cited example of very large daily token counts during a recent Iran operation — illustrate how combat tempo can drive sudden, large costs that are hard to predict.
- The rollback raises short-term budgeting questions about token availability past October 2026 and deepens internal skepticism as the Pentagon expands AI into sensitive workflows and shifts toward consumption-based vendor pricing.