25568ISO 25568 Revision CheckerDraft Technical Specification, first edition Stage 50.00 · 29 Sep 2026

Home / Requirement record / A.9

A.9

Robustness

ISO/IEC 23894:2023 · Annex A: AI-related objectives · annex page · cited to ISO/IEC 23894:2023, clause A.9

AI-related objective: The AI system keeps performing as intended under perturbation, unexpected input, distribution shift and adversarial conditions, and degrades in a controlled way. Used when establishing risk criteria (6.3.4), identifying consequences (6.4.2.6) and evaluating risk (6.4.4). The artefact behind the answer is robustness test results. Read robustness against the objective the use threatens most.

What it asks
AI-related objective: The AI system keeps performing as intended under perturbation, unexpected input, distribution shift and adversarial conditions, and degrades in a controlled way. Used when establishing risk criteria (6.3.4), identifying consequences (6.4.2.6) and evaluating risk (6.4.4).
Where generative AI changes the answer
Robustness is tested by an input designed to break the system, not by the demo that was built to show it working.
Reading it in practice
Robustness is tested by an input designed to break the system, not by the demonstration built to show it working. The evidence is an adversarial test record that includes the failures, the fix and the re-test. A record with no failures is usually a record of a demonstration.
Who it binds
The AI risk owner, with the objective owner named in the risk treatment plan. For A.9 the named role is the owner of robustness test results.
Evidence an auditor normally samples
  • Robustness test results
The test you can run
Take robustness test results and trace it to the decision it supported; if the trail stops before a named person, the control is not evidenced.
Common failures
  • Performance measured on clean test data only
Before the draft publishes
For A.9, pick the two objectives the use most affects, write the test you would apply, and record the result before the draft is published. The first artefact to produce is robustness test results.
Where this sits against the draft
A.9 carries the annex a: ai-related objectives part of the draft's published scope, which names risk sources, the stakeholders exposed to them, risk analysis, risk treatment and controls. When the draft publishes, compare its wording on A.9 (robustness) with this record and note the change in the register.