PII Minimization Handbook (Patricia Thaine, Kathrin Gardhouse) (z-library.sk, 1lib.sk, z-lib.sk)

【One-Line Pitch】 A practical field guide for privacy, compliance, and data teams who need to collect and use less personal data without losing business value—covering the legal principles, the standards, and the day-to-day engineering and cultural changes that make minimization real. 【Book Arc】 - **Opening (~0%–10%)**: Frames the core vocabulary—PII as direct and indirect identifiers, and the three linked concepts of data minimization, de-identification, and anonymization—and states the book's ambition to close the theory-to-practice gap and debunk myths about anonymization in AI and Big Data. - **Early (~10%–30%)**: Surveys how data minimization is expressed across jurisdictions and regimes, including GDPR-style purpose limitation and consent conditions, HIPAA's Safe Harbor versus Expert Determination, and the differing treatment of pseudonymization and anonymized data in laws such as the LGPD, APPI, DPDPA, and Singapore's PDPA. - **Early–Middle (~30%–40%)**: Moves from law to standards, using ISO/IEC 27559:2022 as a lifecycle framework (context, data, identifiability assessment, governance, post-sharing reassessment) and noting that PCI DSS implies minimization through access and retention limits without defining necessity. - **Middle (~40%–50%)**: Applies minimization across the AI system lifecycle—pre-processing, training, and deployment—covering batching/streaming, adaptive sampling, and regulatory signals such as the EDPB's Opinion 28/2024 on AI model processing. - **Late (~50%–90%)**: Shifts to organizational practice: balancing compliance with business goals, the marginal utility of data, alternative data models, and building a minimization culture through training, documentation, and internal marketing. - **Ending (~90%–100%)**: Closes with detailed financial-sector use cases (customer onboarding, credit assessment, fraud prevention, AML monitoring) and forward-looking trends in technology, regulation, business models, and cross-border challenges. 【Key Takeaways】 - **Minimization is purpose-bound, not absolute** (Early): The principle only makes sense relative to a stated purpose, and it works alongside purpose limitation and consent rules—so teams must define purposes before deciding what data is "necessary." - **Consent alone is not a safe harbor** (Early): Some jurisdictions require processing to be adequate, legitimate, and appropriate even with consent, meaning over-collection can be unlawful regardless of permission. - **Anonymization and de-identification are distinct, and definitions vary by law** (Early): HIPAA, GDPR, and Singapore's PDPA treat reversibility and identifiability differently, so a dataset "anonymized" under one regime may still be personal data under another. - **Safe Harbor is simple but incomplete** (Early): Removing the 18 HIPAA identifiers can miss longitudinal or combinable data that raises re-identification risk, which is why Expert Determination—using suppression, generalization, and perturbation—is often preferred for health data. - **Standards provide the operational scaffolding** (Early–Middle): ISO/IEC 27559:2022 supplies a repeatable lifecycle for de-identification governance, while PCI DSS shows how retention and access limits function as de facto minimization without clear necessity guidance. - **Minimization belongs inside the AI lifecycle** (Middle): Limiting data flow during training via batching, streaming, and adaptive sampling reduces exposure and can improve model efficiency—minimization and performance are not inherently opposed. - **Culture and incentives determine adoption** (Late): Business-compliance balance, marginal utility of data, and internal marketing of minimization matter as much as technical controls; habits and daily decisions shape whether policy sticks. - **Sector use cases make it concrete** (Ending): Financial-services scenarios—onboarding, credit, fraud, AML—show how minimization trade-offs play out where regulatory obligations and operational needs collide. 【Reading Tips】 - Deep-read the opening chapters on PII classification and the minimization/de-identification/anonymization distinction; skim the jurisdictional survey if you already know your own regime, but note where definitions diverge. - Treat the ISO/IEC 27559 and HIPAA Expert Determination sections as reference material to return to when designing a de-identification process. - For AI teams, focus on the training-stage techniques (batching, adaptive sampling) and the EDPB discussion; the excerpts do not cover deployment-stage details in depth. - Read the financial use cases last as synthesis—they reward having absorbed the legal and standards material first. - Keep the recurring "Key Takeaways" boxes as checkpoints; they distill each chapter's argument. 【Coverage Limits】 This guide is based on stratified excerpts covering roughly the first half of the book plus the table of contents and closing sections; detailed content of the late business-culture and use-case chapters is inferred from headings and summaries rather than full text.