Article 10: Data governance
Article 10 addresses data and data-governance practices for certain high-risk AI systems. Confirm the applicable requirements with qualified legal counsel.
Start by identifying the datasets used by the AI system. VeriProof receives only the application records your integration sends.
It does not inspect a model provider’s complete training, validation, or test datasets. Collect that evidence from the responsible provider or dataset owner.
Supporting records
- application and environment context;
- recorded model, dataset, or version references;
- captured quality and provenance fields;
- policy results and evidence gaps;
- review and remediation history; and
- scoped evidence exports.
Review the data evidence
- Identify the dataset, source, purpose, owner, and affected application.
- Define the quality, relevance, representation, and bias methods used by the organization.
- Confirm which results VeriProof receives.
- Review missing or conflicting provenance.
- Record the conclusion and corrective action outside VeriProof when the product record is insufficient.
Related guidance
Last updated on