An unnamed enterprise reportedly incurred a half-billion-dollar bill on Anthropic's Claude after failing to set usage limits for employees, while Uber engineers reportedly exhausted their entire 2026 AI budget prematurely. AI agents, unconstrained by guardrails, could exacerbate these token spend disasters, as they generate both input and output tokens that are processed by expensive GPU servers. OpenAI and Anthropic charge users on a token basis, with prices varying by model and output tokens often costing more.