Saudi Arabia SDAIA AI Ethics Principles (Version 2.0, 2025) 7.PID: Accountability and Responsibility in Prepare Input Data: data quality checks, approved and signed-off datasets cleansed of societal bias, and documented, validated preparation

What does 7.PID require?

Data quality checks, cleansing and integrity assurance are performed so results are accurate and intended behaviour is captured in supervised and unsupervised models; datasets are approved and signed off before model development begins, cleansed of societal bias with sensitive features excluded or their inclusion explained as a documented trade-off; the preparation process and quality checks are documented and validated by responsible parties; and data is acquired, classified, processed and made accessible in a way that eases later human intervention and control.