A proxy feature is a data point that is not sensitive on its own but can still reveal, approximate, or reinforce a protected attribute when combined with other data. In security and privacy work, proxies matter because systems often do not need to store a sensitive field directly to infer it indirectly. For example, ZIP code, school, device language, browsing patterns, or login time can correlate with location, income, disability, or other protected traits.
Proxy features show up in real attacks when adversaries use seemingly harmless metadata to profile users, deanonymize records, or target victims more precisely. They also appear in defensive systems and machine-learning models, where weak feature selection can create biased decisions or unintended disclosure. Good defenses include data minimization, correlation reviews, feature audits, and tests that ask whether a model can reconstruct sensitive attributes from non-sensitive inputs.