One Company Spent Half a Billion Dollars on AI in a Single Month. Nobody Set a Limit.

An AI consultant told Axios that one of their enterprise clients ran up a $500 million bill for Claude in a single month. Not a year. A month.

There was no fraud, no breach, no rogue actor. The company handed out AI licenses across its workforce with no usage limits, no spend caps, and no controls over how employees consumed AI resources. Engineers escalated from quick chatbot questions to massive, multi-step automated workflows that ran around the clock. The meter just kept spinning. By the time anyone looked at the invoice, half a billion dollars was gone.

This is what an AI deployment with no human in the loop looks like on the balance sheet. The technology did exactly what it was told. The failure was governance: no gate at deployment, no monitoring of consumption, no one accountable for watching the number climb. Per-developer AI usage has reportedly jumped roughly 18x in nine months as workflows automate — the gap between “we gave everyone access” and “we set the rules” is now a financial cliff.

$500 million is the kind of money that ends careers, kills budgets, and shows up in a 10-K. It evaporated not because AI is dangerous, but because the oversight layer was never built. That is the entire 38 Flags thesis, written in a single invoice.

HITL Score: 24/100 — human oversight at deployment 3/25 (AI access rolled out across the workforce with zero spend caps, zero per-seat limits, zero approval gates; deployment with no human defining the boundary is the textbook governance failure), ongoing monitoring 4/25 (the bill reached $500M before anyone noticed — no real-time consumption dashboard, no spend alert, no threshold trigger), incident response 7/25 (the problem was eventually caught and surfaced by a consultant, after the fact, with no kill-switch on runaway spend), accountability 10/25 (the controls to prevent it — usage limits, budgets, role-based access — plainly exist and simply weren’t switched on).