Definition
An attack that drains an organization's AI budget by flooding a model or agent with high-cost requests until the account is throttled or the bill balloons. Unlike data theft, it attacks availability and cost — holding the service and its spending hostage. It is the AI-era sibling of a denial-of-service attack, aimed at token limits rather than a server.
Why it matters
AI usage bills and rate limits are a real operational-risk plane, and regulators now treat token consumption as a first-order resilience concern. A single hostile or runaway agent can exhaust quota and take a critical AI service offline.