CISA and Canadian Centre for Cyber Security Issue Guidance on Managing AI Token Consumption
CISA and the Canadian Centre for Cyber Security have released joint guidance to help organizations manage the consumption of tokens for AI services, aiming to prevent unpredictable costs and ensure operational resilience.
CISA, in collaboration with the Canadian Centre for Cyber Security (Cyber Centre) and other partners, has issued a new set of guidance titled "Managing AI Token Consumption for Operational Resilience." This advisory aims to equip organizations with the knowledge and strategies necessary to effectively manage their usage of tokens for third-party super intelligence (SI) services.
The rapid adoption of AI services, particularly those with token-based usage models, presents organizations with potential challenges. Uncontrolled token consumption can lead to a cascade of negative consequences, including unpredictable and escalating costs, reduced availability of critical services, degraded performance, and even outright denial-of-tokens or denial-of-wallet attacks. These issues can significantly impact an organization's ability to maintain operational continuity and deliver essential services.
The guidance provides a framework of practical steps designed to mitigate these risks. Organizations are advised to prioritize token allocation for mission-critical functions, optimize their workflows to reduce unnecessary token usage, and improve user prompting practices to ensure more efficient interactions with AI models. Furthermore, the guidance emphasizes the importance of leveraging caching mechanisms and reusable context to avoid redundant computations and data submissions.
Optimizing data before it is submitted to AI models is another key recommendation. By pre-processing and refining data, organizations can reduce the volume of information processed, thereby lowering token consumption and improving the speed and accuracy of AI responses. Continuous monitoring of token consumption over time is also crucial for identifying trends, detecting anomalies, and making informed adjustments to usage policies.
Critical infrastructure owners and operators are specifically encouraged to utilize this guidance to identify which of their functions rely on SI services. Understanding these dependencies, along with the operational dependencies on service providers, is vital for planning and ensuring service availability, especially during periods of high demand, system outages, or security incidents.
Beyond critical infrastructure, the guidance is also valuable for cyber defenders. It offers strategies for managing AI-enabled cyber defense workflows while maintaining essential human oversight, governance, and robust cybersecurity fundamentals. This ensures that AI tools enhance, rather than compromise, security operations.
The overarching goal of this guidance is to foster operational resilience in the face of increasing AI integration. By proactively managing token consumption, organizations can better control costs, ensure the availability and performance of AI-driven services, and protect themselves from potential denial-of-service attacks targeting AI resources.