Control A.6.2.4: AI system verification and validation
Set out in writing how the AI system will be verified and validated and the criteria that govern when each measure is used.
What control A.6.2.4 asks
Set out in writing how the AI system will be verified and validated and the criteria that govern when each measure is used.
Why it matters for privacy
Verification asks whether the system met its requirements; validation asks whether it works for the intended use. A privacy test that only checks accuracy has tested neither.
Where the draft lands: proof before deployment. The abstract asks for risks to be evaluated. A test result is the strongest form of evaluation an organization can hold.
Basis for A.6.2.4: the ISO page 56582 and the ISO/IEC 42001:2023 control record. The overlay is our reading of the published abstract and the stage record, not the draft text. The lifecycle and where privacy risk enters.
Evidence for control A.6.2.4
This is the evidence an auditor or a privacy reviewer asks to see. For control A.6.2.4 the list is written for an AI system that touches personal data, so a system with no personal data will need a subset.
- Verification and validation V&V plan with test methods, tools, representative test data and release criteria
- Verification and validation Evaluation criteria including impacts on individuals and societies, reliability and safety requirements with acceptable error rates, responsible AI objectives and operational factor ranges
- Verification and validation Evaluation results and handling of unmet criteria
Where control A.6.2.4 commonly fails
- Test data not representative of the domain of use
- No acceptable error rates defined
- Failing criteria accepted without reconsidering intended use
What to do about control A.6.2.4 before the draft lands
Add privacy tests to verification and validation: data minimisation, output leakage, prompt injection that reveals records, retention and deletion after the test.
The requirement for control A.6.2.4 is ISO/IEC 42001:2023, and a certificate is issued against that standard or ISO/IEC 27701. The privacy work extends the control to personal data in an AI system; it is not a new system and not a certification.
Read across from control A.6.2.4
- Control A.4.2, Resource documentation
- Control A.4.3, Data resources
- Control A.4.4, Tooling resources
- Control A.4.5, System and computing resources
← Control A.6.2.3 | Control A.6.2.5 →
Open the full index. The public record behind control A.6.2.4 shows the scope and the lifecycle.
The record behind control A.6.2.4 is the ISO/IEC 42001:2023 Annex A set, retrieved 22 October 2026. The public 27091 abstract supplies the privacy direction, not a draft requirement. Check the sources.