Clause 5.5: Robustness
Robustness describes how far the system keeps its level of correctness under difficult conditions: unseen, biased, adversarial or invalid inputs, outside interference, environmental conditions, and the ordinary expectation that the system works as its developers intended.
What clause 5.5 is
Robustness describes how far the system keeps its level of correctness under difficult conditions: unseen, biased, adversarial or invalid inputs, outside interference, environmental conditions, and the ordinary expectation that the system works as its developers intended. Safety and security depend on it. The second edition points to the functional safety work for AI systems, replacing the older technical report reference.
What the 2023 edition said
The first edition introduced robustness as a new sub-characteristic of reliability, listing the same conditions and pointing to the AI trustworthiness and assessment technical reports.
What the second edition changes
The parent stays reliability. The reference for assessment moves to the functional safety series for AI systems. The substance is unchanged.
The FDIS text is not public. The change line above is read from the public draft, which is not the FDIS and may differ. No clause level diff is invented here. The ISO stage record is the authority for the dates.
Why it matters to the buyer
Most failures in production come from inputs the test set did not contain. The clause names the categories of difficult input, which gives a test plan its structure.
What to write in the requirement
Build a robustness test set with a labelled case in each named category: unseen, biased, adversarial and invalid input, plus interference and environment. Record what the system did, not only whether it passed. Repeat after each model change.
Evidence a reviewer asks for
A reviewer will ask which categories were tested and which were skipped. A clean test set is not a robustness result.
- Test setA labelled case in each named category, with the expected and observed behaviour
- ResultsThe robustness results per category, with the failure modes named
- Incident logProduction incidents that show a robustness failure and the fix
Where it commonly fails
- Only clean data was tested
- No adversarial cases
- The test set predates the model
What to do before publication
Add one case per named category before the next release. The list is short and the gap is easy to show.
Read across
Source for clause 5.5: ISO/IEC 25059, Clause 5.5, retrieved 20 October 2026. The change line rests on the public draft of the second edition and ISO page 88234. The requirement, evidence and gaps are written from our record. Every source.