← Graph

Obfuscation and re-identification via record counts / outliers

question 2 connections

Audience member notes obfuscation can still leak identity via counts and outliers (e.g. a caretaker linked to 7 children remains identifiable as the largest family even if individual fields are faked) and asks whether the team randomly deletes/adds records to mitigate this. Sergyenko: they didn't hit this specific leak but did encounter 'excessive data exposure' where therapists entered sensitive info into non-sensitive fields and it leaked into New Relic. Grazer allows custom strategies, so one could write a custom rule that randomly perturbs the number of children per family.

answer_summary
Not a silver bullet — write custom Grazer rules for specific statistical leaks (e.g. randomly perturb number of children per family); related problem is 'excessive data exposure' via free-text fields leaking to third parties like New Relic.
question Obfuscation and re-identification via record counts / outliers
about
About limits of obfuscation against statistical re-identification.
question Obfuscation and re-identification via record counts / outliers
asked_at
Asked during Q&A.

Provenance

Read by
2 extractions