ISO 42001 Annex A 7.4 Quality of data for AI systems
What it says
The organization shall define and document requirements for data quality and ensure that data used to develop and operate the AI system meet those requirements.
Quoted as published. Everything else on this page is written by Control Stack.
What it means
In plain EnglishDefine and document data quality requirements, and ensure the data used to develop and operate the AI system meets them.
Your organisation must define and document requirements for data quality, and ensure that the data used to develop and operate the AI system meets them. Both parts are required:
- Requirements: how complete, accurate, current and representative the data must be, for example. Your organisation decides these, based on the system's purpose and risks.
- Meeting them: the data must actually pass, both in development (training, validation and test data for a model) and in operation (what the running system receives).
Free. Save your status, tasks and notes.
Why it matters
A model built on incomplete or skewed data can give wrong or unfair results, and production data can degrade long after go-live. Written requirements give you a test to apply before and after launch, so bad data can be caught and fixed.
How to implement
List the data used to develop and operate each AI system.
Separate development data (model training data, validation data and test data) from production data, which the running system receives. Start from the data resources documented under Annex A 4.3, then compare them with pipeline and integration settings so nothing is left out.
Write quality requirements for each kind of data.
Choose measures that fit the system's purpose and risks, such as the share of missing values, the accuracy of labels (the answers a model learns from), the age of records, and how well the data represents the people and situations the system will meet. Your organisation sets each pass level.
Test development data before use.
Run validation checks so a failing data set stops the model build. Fix or replace failing data (Annex A 7.6 covers preparation methods). Before use, change a requirement only if it is wrong, not because data failed it, and record why. After model training, compare results across groups, because a difference can reveal skewed data.
Check production data against its requirements while the system runs.
Add input checks for format, range and missing values, and monitor for drift (incoming data moving away from what the model was tested on). Send alerts to a named person, and decide beforehand whether failing input is rejected or the system is paused.
Cover bought AI services through the supplier and your inputs.
Where you cannot test a supplier's training data, its documentation is your evidence: compare your requirements with how that data was chosen and checked. Record gaps, including anything undisclosed, and decide whether to keep the service for this use. Run your own checks on data sets and feeds you supply.
Keep it running
Keep the data checks running, with alerts going to a named person. Look again at the requirements and re-test when a data source changes, the model is retrained, the system's purpose or users change, or monitoring shows drift.
Evidence and audit
- Check 1 of 4What an auditor asks for
Ask for
The data quality requirements for AI systems chosen from software and supplier records, with the pipeline and integration settings showing what data each uses.
Look at
Whether the settings show any data set or feed with no requirement, and whether development data and production data are both covered by requirements specific enough to test.
Good evidence
Every data set and feed found in the settings has a documented requirement specific enough to test, covering both development and operation.
- Check 2 of 4What an auditor asks for
Ask for
Validation results for the data used in the latest model build, plus one check re-run on that stored data.
Look at
Whether each requirement was measured on that data, whether the re-run gives the same answer, and what happened to data that failed.
Good evidence
Each requirement has a measured result for the data actually used, and the re-run agrees. Data that failed was fixed or replaced before use. Otherwise, the requirement was changed before use by a recorded decision giving a reason other than the failure itself, and the data then met the changed requirement.
- Check 3 of 4What an auditor asks for
Ask for
A demonstration of the production data checks, using a test input that breaks a requirement.
Look at
Whether the test input is stopped as decided beforehand, who is told, and whether any input route in the data flow skips the checks.
Good evidence
The test input is rejected or the system is paused, as decided beforehand, the named person receives the alert, and every input route in the system's data flow passes through the checks.
- Check 4 of 4What an auditor asks for
Ask for
The supplier's data documentation and your assessment of it, for each bought AI service in the supplier records.
Look at
Whether the assessment compares the documentation with your requirements, whether each gap has a decision, and whether the data sets and feeds you supply are tested.
Good evidence
Each bought AI service has an assessment against your requirements, each gap, including anything undisclosed, has a recorded decision, and every data set and feed you supply to the service has been tested against your requirements.
Mappings
Closest match in ASD Information Security Manual: ISM-2087, Verify the source and integrity of AI training data, which partially overlaps with this control.
ASD Information Security Manual: 2 mappings
This control partially overlaps ISM-2087.
Annex A 7.4 demands documented data quality requirements to ensure AI development and operational data meets those standards. ISM-2087 requires verifying the source and integrity of AI training data, addressing the provenance and integrity aspects of Annex A 7.4’s framework. Annex A 7.4 extends its focus to additional quality attributes and encompasses all data used in AI operations. The overlap lies in their mutual emphasis on integrity and provenance assurances.
This control partially overlaps ISM-2088.
Annex A 7.4 requires the organisation to define, document and verify data quality requirements for data used to develop and operate AI systems, emphasising dimensions such as accuracy and completeness. ISM-2088 focuses specifically on using data validation techniques to ensure the reliability and accuracy of AI model training data, which aligns with the quality verification activities in Annex A 7.4. While Annex A 7.4 is broader, encompassing various quality dimensions and applications beyond training data, the overlap occurs in the shared goal of ensuring data reliability.
Mappings show how controls relate across frameworks. They do not imply full equivalence or certification.
Get help with this control
ISO 42001 training
Mindset Cyber, a PECB Authorised Titanium Partner, sells official PECB ISO/IEC 42001 courses that map directly to the AI controls in this library.
Turn this control into your AI policy
PolicyMint writes customised ISO 42001 policy documents for your business, audit-ready and exported in your branding (Word, PDF, Excel). Skip the blank page and the boilerplate that does not fit. First document free.
Draft your ISO 42001 documents



