Communications of the ACM
The Anonymization Debate Should Be About Risk, Not Perfection
Credit: Alicia Kubista / Andrij Borys Associates
For years, the key ethic for safe, sustainable data sharing was anonymization. As long as a researcher or organization took steps to anonymize datasets, they could be freely used and shared. This notion was even embedded in law and policy. For example, laws like the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule and the European Union's Data Protection Directive facilitate sharing of anonymized datasets with fewer if any restrictions placed upon datasets that contain personal information.
But it turns out that "anonymization" is not foolproof. The possibility of correctly identifying people and attributes from anonymized datasets has sparked one of the most lively and important debates in privacy law. In the past 20 years, researchers have shown that individuals can be identified in many different datasets once thought to have been fully protected by means of de-identification.a,7 In particular, a trio of well-known cases of re-identification has called into question the validity of the de-identification methods on which privacy law and policy, like the HIPAA privacy rule, relies. A governor and Netflix and AOL customers were all accurately identified from purportedly anonymized data. In each case, an adversary took advantage of auxiliary information to link an individual to a record in the de-identified dataset.
No entries found