An AI token budget is a capped amount of model usage assigned to a team, application, or workflow. It can cover prompts, generated output, and agent actions, and is usually tracked as a cost, quota, or rate limit. In practice, tokens are the units a model counts as it processes text, so a budget limits how much an AI system can think, write, call tools, or loop on a task.
In cyber security, token budgets matter because they control spend, reduce abuse, and limit damage if an agent is compromised. Attackers can waste a budget with prompt flooding, trigger expensive tool chains, or exploit agent loops to create denial-of-wallet conditions. Defenders use budgets to contain runaway automation, separate high-trust security tasks from general business use, and enforce least privilege for AI agents. A tight budget can also reveal suspicious behavior when usage spikes unexpectedly.