Privacy by design

In plain English

Architects design fire safety into a building: exits, sprinklers and fire-resistant materials. They do not wait for a fire and then retrofit. Privacy by design applies the same logic to systems that handle personal data.

Ann Cavoukian, a former Information and Privacy Commissioner of Ontario, developed the idea in the 1990s. Her seven foundational principles are:

1. Proactive not reactive; preventative not remedial
2. Privacy as the default setting
3. Privacy embedded into design
4. Full functionality: positive-sum, not zero-sum
5. End-to-end security: full lifecycle protection
6. Visibility and transparency: keep it open
7. Respect for user privacy: keep it user-centric

"Full functionality" is the principle the exam likes most. It rejects the idea that privacy must cost you a working product. In October 2010, data protection regulators at their international conference unanimously adopted a resolution recognizing privacy by design.

Example

Little Otter Learning, a fictional company, is designing a reading app that listens as children read aloud. At the first design review, before any code is written, the team decides:

• On-device processing. Speech recognition runs on the tablet, and audio never leaves it by default.
• Default off. Sharing recordings to improve the model stays off unless a parent opts in.
• Minimization. The app stores reading scores, not audio, and deletes scores 12 months after an account closes.
• Access. Teachers see their own class only.
• Testing. Before each release, the model is checked for memorized voice data from training.

Children still get instant feedback, and parents who opt in help improve accuracy. Privacy and the product both win.

Why it matters

• GDPR Art. 25. Data protection by design (Art. 25(1)) requires appropriate measures, such as pseudonymization, both when deciding how to process and during processing. Data protection by default (Art. 25(2)) limits collection, processing, storage time and accessibility to what each purpose needs (GDPR). The EDPB Guidelines 4/2019 explain both duties.
• Fines. Breaching Art. 25 can bring fines of up to €10 million or 2% of worldwide annual turnover, whichever is higher (Art. 83(4)(a)).
• AI design choices. Privacy by design shapes architecture, such as federated learning, differential privacy and on-device processing. It also covers prompt and log retention, and whether user data trains models by default.
• NIST. The AI RMF says privacy values such as anonymity, confidentiality and control should guide AI design choices.

How the AIGP exam tests it

Lesson II.A.2 pairs privacy by design with data minimization.

• The tell. Proactive, embedded and on by default. The right option settles privacy at the design stage.
• Trap 1: fix it in version 2. Launching first and adding privacy controls after complaints is reactive, the opposite of privacy by design.
• Trap 2: a setting three menus deep. Protection that users must switch on fails "privacy as the default."
• Trap 3: privacy vs functionality. Options that frame privacy as a cost to the product miss "positive-sum, not zero-sum."
• By design or by default? Pseudonymizing data before training is by design. Logs that auto-delete unless someone changes the setting are by default.