- CISA, Canada’s Cyber Centre and partners released joint guidance on managing AI token usage.
- Unmanaged token consumption may contribute to higher costs, reduced availability, degraded performance, token exhaustion and denial-of-tokens or denial-of-wallet attacks.
- Critical infrastructure operators can use the guidance to identify mission-critical functions that depend on super intelligence services and plan for reduced availability.
CISA, the Canadian Centre for Cyber Security and partners released joint guidance on managing AI token consumption on Oct 22, 2026. The guidance is intended to help organizations manage token usage in support of operations.
CISA said token-based usage models can affect cost, performance and access to critical services as organizations adopt third-party super intelligence services. Unmanaged consumption may contribute to rising and unpredictable costs, reduced service availability, degraded performance, token exhaustion, and denial-of-tokens or denial-of-wallet attacks.
The guidance also addresses operational dependencies. Critical infrastructure owners and operators can use it to identify mission-critical functions that depend on super intelligence services and plan for reduced availability during outages, incidents or periods of high demand.
On this page
Agentic workflows can multiply token consumption
Ajith Sankaran said enterprise AI spending is volatile, opaque and difficult to govern as organizations move from experimentation to scaled deployment. He said agentic workflows can multiply token consumption when models retrieve documents, reason through tasks, invoke tools and generate iterative outputs.
Sankaran recommended centralizing monitoring of model usage, token consumption, inference workloads and spending by business unit. He also recommended setting policies for model access, agent autonomy, API usage and multimodal workloads before consumption patterns become entrenched.
Vendor, CISA and Deloitte views on AI adoption and spend control
CloudEagle.ai said finance and IT can set a spending threshold for each team or application, monitor usage in real time and cut it off automatically.
CISA said cyber defenders can use the guidance to manage super-intelligence-enabled cyber defense workflows while maintaining human oversight, governance and strong cybersecurity fundamentals.
What to do now
- Prioritize token allocation, design efficient workflows, improve prompting, use caching and reusable context, optimize data before submission, and monitor token consumption over time.
FAQ
What did CISA release on Oct 22, 2026?
CISA, the Canadian Centre for Cyber Security and partners released joint guidance on managing AI token consumption for operational resilience.
Who should use the AI token consumption guidance?
Organizations adopting third-party super intelligence services can use it. Critical infrastructure owners and operators can use it to identify mission-critical functions that depend on those services, while cyber defenders can use it for super-intelligence-enabled defense workflows.
What can happen when AI token consumption is unmanaged?
CISA said it may contribute to rising and unpredictable costs, reduced service availability, degraded performance, token exhaustion, and denial-of-tokens or denial-of-wallet attacks.
How can organizations manage AI token usage?
The guidance outlines prioritizing token allocation, designing efficient workflows, improving prompting, using caching and reusable context, optimizing data before submission, and monitoring consumption over time.
Why can agentic AI increase token use?
Ajith Sankaran said token consumption can multiply quickly when multiple models autonomously retrieve documents, reason through tasks, invoke tools and generate iterative outputs.
Sources
- Managing AI Token Consumption for Operational Resilience, CISAPrimary
- Deloitte report: AI is shifting from pilots and experimentation to scaling, DeloittePrimary
- The Economics Of Generative AI: Why Managing Token Costs Is A Strategic Imperative, Forbes
- CloudEagle.ai Gives Enterprises Complete Control Over AI Token Consumption and Budgeting, Amarillo Globe-News
How we checked this story
| Claim | Source | Status |
|---|---|---|
| CISA, the Canadian Centre for Cyber Security and partners released joint guidance to help organizations manage token usage. | CISA | Confirmed |
| CISA said token-based usage models can affect cost, performance and access to critical services as organizations adopt third-party super intelligence services. | CISA | Confirmed |
| CISA said unmanaged token consumption may contribute to rising and unpredictable costs, reduced service availability, degraded performance, token exhaustion and denial-of-tokens or denial-of-wallet attacks. | CISA | Confirmed |
| CISA said critical infrastructure owners and operators can use the guidance to identify mission-critical functions that depend on super intelligence services. | CISA | Confirmed |
| CISA said cyber defenders can manage super-intelligence-enabled cyber defense workflows while maintaining human oversight, governance and strong cybersecurity fundamentals. | CISA | Confirmed |
| Ajith Sankaran said enterprise AI spending is volatile, opaque and difficult to govern as organizations move from experimentation to scaled deployment. | Ajith Sankaran, via Forbes | Attributed |
| Ajith Sankaran said agentic workflows can cause token consumption to multiply quickly when models retrieve documents, reason, invoke tools and generate iterative outputs. | Ajith Sankaran, via Forbes | Attributed |
| Ajith Sankaran recommended centralized monitoring of model usage, token consumption, inference workloads and spending by business unit. | Ajith Sankaran, via Forbes | Attributed |
| Ajith Sankaran recommended establishing policies around model access, agent autonomy, API usage and multimodal workloads before consumption patterns become entrenched. | Ajith Sankaran, via Forbes | Attributed |
| CloudEagle.ai said finance and IT can set a spend threshold per team or app, monitor usage in real time and automatically cut it off. | CloudEagle.ai, via Amarillo Globe-News | Attributed |
| Deloitte said companies broadened workforce access to AI by 50% in one year, with a majority of workers equipped with sanctioned AI tools. | Deloitte | Confirmed |
| Deloitte said close to three-quarters of companies planned to deploy agentic AI within two years, while only 21% reported a mature agent-governance model. | Deloitte | Confirmed |
| CISA published its report on Oct 22, 2026. | CISA | Confirmed |
| deloitte.com published its report on Jan 21, 2026. | deloitte.com | Confirmed |
Could not verify
- Whether denial-of-tokens or denial-of-wallet attacks have occurred in the wild is not established
- How many organizations or services are affected is not established
- Whether the guidance addresses a specific incident or vulnerability is not established
- Whether the guidance assigns a CVE or CVSS score is not established
Related stories
AI-Powered AttacksOperation Serengeti III arrests 1,462 suspects across 21 countries
AI Supply ChainRank Math PRO malware could compromise WordPress sites
Vulns & PatchesSiemens Industrial Edge flaws could expose sensitive information