⚠️
AI Privacy Risk
Privacy
Risk category covering inference, re-identification, memorization, and sensitive attribute extraction.
Definition
AI privacy risk includes: inference of sensitive attributes, re-identification from “anonymous” data, model memorization of training examples, and leakage via logs or outputs.
In plain English
Risk category covering inference, re-identification, memorization, and sensitive attribute extraction.
Why this matters
Why it matters: Automated inference can bypass user intent and increase discrimination and surveillance.
Example
Example: Run privacy impact assessments, restrict sensitive features, and limit retention and sharing of training and inference logs.