Which practice supports auditability in AI systems?

• A. Model performance metrics alone are sufficient for auditability.
• B. Data lineage and transformation records across lifecycle.
• C. Ignoring logs.
• D. Comprehensive documentation of data sources, transformations, and decisions supports auditability.

Answer: (D)

Auditability in AI systems hinges on having a complete, verifiable record of where data comes from, how it is transformed, and why the model makes a given decision. Comprehensive documentation of data sources, transformations, and decisions provides that verifiable trail, enabling stakeholders to trace inputs to outputs, reproduce results, and perform forensics or compliance reviews across the lifecycle. Model performance metrics alone may show how well the model performs but don’t reveal the data processing path or the reasoning behind specific outputs. Relying only on data lineage and transformation records captures some aspects of traceability but without documenting the decision context and data provenance in a unified way, auditability remains incomplete. Ignoring logs eliminates critical evidence and breaks the trail entirely.