Data anonymization removes personal identifiers so the data cannot be linked back to individuals. Names, addresses, phone numbers, and social security numbers get stripped or replaced. The goal is to use the data for research or analysis without exposing anyone's identity. Healthcare researchers anonymize patient records. Companies anonymize user behavior data before sharing it with partners.

True anonymization is harder than it sounds. Removing names is not enough. Combinations of attributes can re-identify people. Birth date, ZIP code, and gender together uniquely identify most individuals in the United States. Latanya Sweeney demonstrated this in 1997 by re-identifying the medical records of a state governor from supposedly anonymous data. The Netflix Prize dataset was de-anonymized in 2008 by cross-referencing with IMDb ratings. Techniques like k-anonymity, differential privacy, and synthetic data generation try to solve the problem. Each has trade-offs between privacy and utility. Stronger anonymization means less useful data. Perfect anonymization may mean no data at all. Regulators recognize the difficulty. GDPR treats anonymized data differently from pseudonymized data, but the line between them is blurry. Re-identification is often just a matter of effort.

Anonymization techniques

• Suppression — remove identifiers entirely
• Generalization — replace specific values with ranges
• Pseudonymization — swap identifiers for fake ones, reversible with a key
• Differential privacy — add statistical noise to protect individuals
• Synthetic data — generate artificial records that mimic real patterns

Anonymization reduces risk. It does not eliminate it. Treat anonymized data with care.