What is an inference attack, and how does it differ from a linkage attack?

An inference attack deduces a sensitive attribute from group membership using the data's internal patterns — no external dataset needed, unlike linkage.

Inference attacks exploit the relationship between group membership and sensitive attributes. Even if the attacker can't identify a specific person, they can infer the sensitive value with high confidence based on group characteristics, using statistical methods or machine learning to find subtle correlations.

The key difference from linkage: inference attacks exploit the internal structure of the data itself rather than relying on an external auxiliary dataset.

Classic failure: a dataset where everyone aged 25–30 in ZIP 94101 has "stomach disease." An attacker who merely knows someone falls in that group learns their medical condition with 100% certainty — regardless of k-anonymity guarantees and without ever identifying them.

Tip: Linkage asks "who is this record?"; inference asks "what's true of anyone in this group?" — and often the second is all the attacker wanted.