Stockertim Dictionary · Artificial intelligence
Inference (privacy)
Also called inferred data
In short
In privacy, inference means deriving new and possibly sensitive information about a person, such as health or beliefs, from other data about them.
- Intermediate
- Concept
- Checked
In detail
Inference, in the privacy sense, is the step from what is known about a person to something new. Shopping habits can suggest a pregnancy, typing patterns can hint at a health condition, and the pages someone reads can point to their politics or religion. AI systems are very good at spotting such patterns, and they can make these guesses without anyone asking them to.
It works because data that looks harmless on its own becomes revealing in combination. A model trained to predict which customers will respond to an offer may, as a side effect, learn signals that track age, disability or financial stress. The inferred trait may be wrong, which causes its own harm, or right, which can expose something the person never chose to share.
Inference matters because privacy is not only about what people hand over. In many data protection regimes, information inferred about a person counts as personal data, and inferences about health, ethnicity, beliefs or sexuality may fall into special categories with stricter rules. Rules differ by country, but the ethical point holds widely: people should not be surprised by what an organisation has worked out about them.
At work, teams ask what a system could infer, not only what it is given, and limit uses that guess sensitive traits without a clear, lawful and fair reason.
Worked example
Lagoon Logistics plans to analyse delivery drivers' phone data to improve routes.
- The team notices that stop times and locations could reveal regular visits to a clinic or a place of worship.
- It asks what the routing model needs: road speeds and delivery times, not personal patterns.
- It aggregates data by route, removes driver identities from the analysis and deletes raw location trails after a short period.
- It tells drivers what is collected and why.
The routing goal is met without building a picture of drivers' private lives.
In practice
Privacy teams add a question about inference to impact assessments, test models for signals that track sensitive traits and apply data minimisation. Staff using AI assistants avoid asking tools to guess a named person's health, beliefs or other private characteristics.
Watch out
- This is the privacy sense of the word; in machine learning, inference also means simply running a trained model to get an answer.
- Removing a sensitive field does not stop a model inferring it from other data.
Learn it properly
Official sources
- Artificial intelligence · Information Commissioner's Office
- A guide to the data protection principles · Information Commissioner's Office
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.