Skip to content
Brief Security & Trust ·

CISA and Canada publish guidance on managing AI token use

The joint guidance covers cost, availability and attack risks for organisations that use token-based AI services.

In brief
  • CISA, Canada's Cyber Centre and partners released joint guidance on managing token use in third-party AI services.
  • Token billing makes AI a variable cost that can be attacked, so budget owners should treat it as a dependency.

CISA said on 22 October that it, the Canadian Centre for Cyber Security and partners issued joint guidance. It is called Managing AI Token Consumption for Operational Resilience. It is for organisations that use third-party AI services, which CISA calls super intelligence services. These services bill by tokens. CISA lists six practices: set priorities for token use, design efficient workflows, improve prompts, reuse saved context, trim data before sending it, and track use over time.

CISA says unmanaged use may bring rising, unpredictable costs, lower availability and slower performance. It also names token exhaustion, denial-of-tokens and denial-of-wallet attacks. As general background, a token is a small unit of text that a model reads or writes. Fees and limits grow with token counts. Long prompts, big files and automated loops use more. That is how a budget or quota can be drained, on purpose or by accident. This brief rests on CISA's announcement page. The paper itself has the detail.

The point for budget owners is that token billing makes AI a variable cost that attackers can target. It is a dependency as well as a cost line. Ask three things. Who can spend tokens? What caps exist? Which functions fail if the service stops?

A WebPulse Brief: a short report of an important event, written by the WebPulse Newsroom with AI assistance and checked against the reporting below. When there is more to explain, we follow up with a full story. How we use AI.

Reporting: CISA.