reidentification

Reidentification, or re-identification, is the process of determining the identity of individuals from data that has been de-identified or anonymized. It often occurs when anonymized data are linked with other data sources containing identifying information, enabling the inference of who is represented in the data.

Risk factors include the presence of quasi-identifiers—data fields that are not unique on their own but can become identifying when combined (for example, birth year, ZIP code, and gender). The increasing availability of public and commercial datasets raises the likelihood of successful reidentification through data linkage and background information.

Historical cases have highlighted the risk. In 1997, Latanya Sweeney showed that Massachusetts voter registration data could be linked with a public governor's name, effectively deanonymizing medical records when used with other attributes. The 2007 Netflix Prize released anonymized viewing data that researchers linked to IMDb profiles to identify some users, illustrating the vulnerability of ostensibly anonymized datasets. The 2006 AOL query log release similarly exposed user identities through search queries and timestamps.

Mitigation approaches include reducing identifiability (data minimization), applying k-anonymity or l-diversity, and employing differential privacy. More robust strategies rely on synthetic data, secure data enclaves, and strict access controls. Privacy-by-design and regular risk assessment are recommended to balance data utility with protection.

Reidentification remains a central concern in data governance and privacy law. Regulations such as the GDPR and HIPAA establish standards for de-identification and govern how data can be shared, while institutions implement technical and organizational safeguards to limit reidentification risk.