Verification and validation test the model on data it has not seen and then check the whole system against the needs of its stakeholders.
What the base framework establishes
The clause separates two questions. Model evaluation compares predictions on test data with the actual labels and reports the metrics the tools group defines. It warns that comparing many models on test data overfits the hyperparameters to the test set, so comparative evaluation should be limited to a few genuinely different candidates. System validation is a different check: whether the trained and deployed model meets stakeholder needs and whether the assumptions about the environment behind the features were correct. The clause says explicitly that system validation is not the same as validation data.
Why it matters to an AI management system
Most teams do the first check and call it validation. The second question, whether the system works in the world it was built for, is the one a board and a regulator actually care about. The clause gives the governance lead a named second test to require.
The public draft summary names no change to 8.5 Verification and validation. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Model evaluation report | Test metrics, the baseline and the comparison set. |
| System validation record | The stakeholder need tested, the method and the result. |
| Environment assumption note | The assumptions behind the features and how they were checked. |
One question to prepare for. What did you test beyond the model's accuracy on held-out data?
Common gaps
- Model metrics are presented as system validation.
- Many models are compared on the test set and the best result is reported.
- The environment assumptions are never tested after deployment.
Prepare before publication
Split the evidence into model evaluation and system validation. Add an environment assumption note and test at least one assumption with real users or real conditions.
Dates that matter
No separate date attaches to 8.5; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
8.4 Modelling · 8.6 Model deployment · 8.7 Operation
Source
Source for 8.5: ISO/IEC 23053:2022, clause 8.5. The public sample stops before 8.5, so this page relies on the public contents and the public summary and does not quote the clause. The two official records are linked on this page so any line can be checked.
Not legal advice; does not certify. A reading of the framework and the evidence a reviewer commonly samples, not a substitute for the standard or your auditor's rules.