What is the concept of 'data minimization' as applied to AI feature engineering, and why does it improve both privacy and security?

• A.A. Data minimization means using the smallest possible training dataset; smaller datasets always produce better-performing models.
• B.B. Data minimization requires collecting and retaining only the minimum personal data necessary for the specified AI purpose: using aggregated or derived features (e.g., 'transaction velocity' instead of individual transaction records) rather than raw personal data reduces re-identification risk, limits breach exposure, and simplifies compliance with data subject rights. From a security perspective, minimizing PII in model training also limits what can be memorized and extracted.
• C.C. Data minimization is a storage cost optimization principle with no privacy implications.