ASD ISM 2090 Rate limiting for AI inference queries
What it says
Rate limiting is applied to inference queries for AI models.
Quoted as published. Everything else on this page is written by Control Stack.
What it means
In plain EnglishLimit the rate at which AI models can be queried to prevent overuse.
This control is about setting limits on how often people can use AI tools to make sure they aren't overused and possibly hacked. If too many queries are allowed, it could lead to system overload or unwanted access, which can disrupt business operations and expose sensitive data.
Free. Save your status, tasks and notes.
Why it matters
Not applying rate limits on AI queries can lead to system overload or security breaches, risking business disruption and data exposure.
How to implement
IT team should set limits on AI service access.
They can do this by configuring software to accept only a set number of queries every hour. This reduces the risk of system overload and possible misuse.
System managers should monitor usage patterns.
They need to regularly check reports for any unusual spikes in AI queries. This helps identify and address potential misuse early.
Business owners should communicate these limits to the team.
They can hold a meeting or send out an email explaining how and why rate limiting is important for protecting the business.
Security personnel should audit the system for bypass attempts.
Regularly review logs for signs of attempts to exceed query limits to catch and respond to malicious activities quickly.
The IT department should continuously update rate limits based on capacity and usage trends.
Evaluate system performance and adjust limits to balance usability with security effectively.
Keep it running
Regularly review and adjust rate limits to match system capabilities and user needs, ensuring both operational efficiency and security.
Evidence and audit
- Check 1 of 5What an auditor asks for
Ask for
the rate limiting configuration document: Request details about the settings controlling AI query limits. Look to ensure limits align with capacity and security needs.
Good evidence
includes documented thresholds tailored to system capability.
- Check 2 of 5What an auditor asks for
Ask for
reports on AI query usage: Request logs showing recent AI queries and any exceeded limits.
Look at
consistency with set policies and spike identification.
Good evidence
shows acceptable usage patterns with logged exceptions.
- Check 3 of 5What an auditor asks for
Ask for
system alert documentation: Request information on alerts configured for when limits are exceeded.
Look at
active alerts that notify the right personnel promptly.
Good evidence
includes timely alerts connected to a response plan.
- Check 4 of 5What an auditor asks for
Ask for
staff meeting notes or emails about rates: Request records of communications regarding rate limits shared with relevant staff.
Look at
clear instructions and rationale.
Good evidence
shows comprehensive communication to involved teams.
- Check 5 of 5What an auditor asks for
Ask for
incident response records: Request any reports or logs from incidents related to exceeding AI query limits.
Look at
appropriate handling and quick resolution of such incidents.
Good evidence
describes prompt investigation and mitigation.
Mappings
Closest match in ISO/IEC 27001: A.8.6, Capacity management for resource use, which this control depends on.
ISO/IEC 27001: 1 mapping
This control depends on A.8.6.
Annex A 8.6 requires monitoring and adjustment of resource use to prevent performance degradation or failures due to capacity shortfalls. ISM-2090 supports this by applying rate limiting to AI model inference queries, which directly reduces the risk of resource exhaustion and helps keep service demand within manageable capacity. ISM-2090 is a targeted protective control for a specific workload, whereas Annex A 8.6 is broader and includes monitoring and planning for overall resource capacity.
Mappings show how controls relate across frameworks. They do not imply full equivalence or certification.