The Importance of Engineering Privacy From the Get Go

The talk opens with a hypothetical food app that quietly shares location, diet and health-related habits with third parties, to show how seemingly harmless data can become harmful. The speaker separates privacy, which is an individual's control over their own information, from security, which is the set of controls that protect it. Personally identifiable information is split into sensitive and non-sensitive categories, and the speaker notes that the split depends on jurisdiction.

Next, the talk walks through de-identification techniques: redaction, masking or pseudonymization, generalization, and noise-based obfuscation. It shows why these fail in practice using two cases: a movie-rating dataset that was linked back to public reviews, and a study that matched released hospital records to newspaper articles. Regulatory fines under GDPR and the patchwork of US state privacy laws are presented as business reasons to act.

To close, the speaker offers five practical rules: collect and store less data, automatically delete old data through a retention policy, process only what a job needs, build privacy and security in from the beginning, and consult a privacy lawyer because the rules change quickly. The takeaway is that engineers who handle user data are its stewards and should design with that responsibility in mind.