Saudi Arabia SDAIA AI Ethics Principles (Version 2.0, 2025) 1.PID: Fairness in Prepare Input Data: representative, accurate data of sufficient size, with sensitive attributes and their proxies excluded or justified

What does 1.PID require?

Acquiring, handling, classifying and managing data responsibly ensures source integrity and accuracy across all observations so that under-represented groups are not systematically disadvantaged nor over-represented groups advantaged, with data quantity and quality sufficient for the purpose since sample size drives accuracy and fairness. The sensitive attributes defined at design, and their proxies, are analysed and kept out of the model data; where accuracy or the system's objective makes that impossible, their use is justified in writing.