Privacy-by-design

As explained in Introduction to Generative AI, Second Edition

Privacy by design is a development approach that integrates privacy protections, proactive risk management, and data minimization throughout a technology's lifecycle.

Privacy by design treats privacy as a core design requirement rather than a fix added after deployment. It identifies risks early, minimizes data use, and applies protective defaults throughout a technology’s lifecycle. For large language models (LLMs), this means addressing privacy from training-data preparation through deployment and later evaluation. section 3.4

A privacy-by-design approach for LLMs can include:

• Curating training data to remove sensitive information. section 3.4
• Using privacy-enhancing techniques during preprocessing and training. section 3.4
• Evaluating models for unintended memorization and documenting the privacy measures used. section 3.4
• Setting privacy protections by default and treating accountability as an ongoing responsibility. section 3.4 section 3.14

Privacy by design is also a legal obligation in the European Union under Article 25 of the General Data Protection Regulation (GDPR), which requires data protection by design and by default. Compliance expectations differ internationally: Brazil’s LGPD and China’s PIPL emphasize related protections, while the United States has traditionally relied more on sector-specific notice and choice. section 3.4 section 3.14

Figure 3.4 A timeline of data protection laws

Browse the timeline to place privacy by design within the development of modern data-protection laws. section 3.14

How Introduction to Generative AI, Second Edition covers it

How deep it goes

Explains lifecycle principles and concrete model-development measures, but does not descend into implementation details.

Examples

Uses training-data curation, preprocessing, training, memorization-risk evaluation, privacy documentation, and default privacy settings as examples.

What you can do after

Apply privacy checks across a model lifecycle, including data curation, protective techniques, evaluation, documentation, and compliance planning.

How it connects

Connects data minimization, privacy-enhancing techniques, memorization evaluation, documentation, GDPR, and data protection by design and by default.

liveBookmanning.com

Read Introduction to Generative AI, Second Edition on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  The Art of AI Product DevelopmentPrivacy-by-design is an approach that embeds privacy protections into the architecture and operation of systems from the outset.
• 

  liveBookmanning.com
  First-Party Data ActivationA design principle that incorporates privacy protections into systems, processes, and products from their inception rather than adding them afterward.

Compare how 3 books explain privacy-by-design →

Related concepts

• data minimization strategyuses · Data minimization is the practice of collecting, processing, using, and retaining only the personal data necessary for specified purposes.
• validationuses · The process of checking data, behavior, systems, models, or results against defined criteria to assess their correctness, suitability, or acceptability.
• principlekind of · A principle is a general rule, guideline, or value that guides decisions, behavior, or judgments.
• federated learningused by · Federated learning is a machine-learning approach in which a shared model is trained across decentralized data sources without centrally collecting their raw data.
• conversational aiconfigures · Conversational AI is artificial intelligence that enables software systems to interact with users through natural-language dialogue.
• privacyconfigures · Privacy is the protection and controlled handling of information or activities to prevent inappropriate access, collection, use, or disclosure.

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