liveBook · Manning

Privacy-by-design

As explained in AI Governance

Privacy by design is an approach that builds privacy protections into the design and operation of systems from the outset.

Privacy by design shifts privacy work to the start of system development. Retrofitting protections after deployment can be expensive; privacy requires more than encryption or access control. section 5.7

Build privacy into the system and verify protections in operation:

  • Minimize data, design for verifiable deletion, and validate outputs from day one. section 5.7
  • Use threat modeling to examine risks such as data poisoning, model inversion, and jailbreaking; refine designs with privacy engineers. section 1.5
  • Apply controls such as AI firewalls, detailed logs, and customer-managed encryption keys for data held by vendors; document capabilities, risks, and mitigations. section 1.5
  • Test retention settings and runtime filters, and preserve evidence such as deletion-job logs, audit trails, and red-team reports. section 5.7 section 1.5
Table 1.3 Overview of the types of tools available at each level
Governance Level
Procedural Tools
Technical Tools
Cultural/Organizational Tools
Strategy & Policy
Responsible AI Charters, Policy Workshops, Risk Tiers
N/A
Cross-functional committees, Accountability mapping
Risk & Impact Assessment
Regulatory Checklists, Impact Assessment Templates
Risk assessment questionnaires and software catalogs
Cross-functional decision bodies
Implementation Review
Threat Modeling Sessions, Design Reviews
AI Firewalls, Customer-Managed Encryption Keys (CMEK), Model Cards
“Privacy by Design” training, Red-teaming exercises
Acceptance Testing
Test Plans, Go/No-Go Criteria
Promptfoo[55], Garak[56], PyRIT[57], Adversarial Robustness Toolbox (ART)[58]
External expert reviews, Simulation workshops
Operations & Monitoring
Incident Response Playbooks, Monitoring Policies
Real-time output scanners[59],[60],[61],[62], Drift-detection dashboards, MLFlow Tracking[63]
Post-incident reviews, Incident escalation paths
Learning & Improvement
Feedback Loops, Policy Update Protocols
Trust score dashboards, Calibration metrics
Culture of transparency, Lessons learned sessions

Use the governance-level overview to locate privacy-by-design practices within implementation review and compare them with testing and monitoring tools. section 1.5

How AI Governance covers it

How deep it goes
Specifies data minimization, verifiable deletion, output validation, and review controls, but stops short of detailed implementation recipes.
Examples
Uses data poisoning, model inversion, and jailbreaking as threat scenarios, and deletion logs and audit trails as evidence.
What you can do after
Apply privacy controls early and verify deletion, retention settings, and runtime filters with operational evidence.
How it connects
Connects privacy with AI threat modeling, data minimization, output validation, deletion, and encryption.

Also explained in

Compare how 4 books explain privacy-by-design →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.