Linkage Attacks and Defenses
What you'll learn: How attackers exploit multiple datasets to pierce through anonymization and what defenses prevent re-identification.
The Core Threat
Even when you've applied k-anonymity, generalization, or pseudonymization to a dataset, attackers don't work in a vacuum. Linkage attacks combine your "anonymized" dataset with external data sources to re-identify individuals. Think of it like solving a puzzle: one dataset alone might hide identities, but when you overlay it with publicly available information (voting records, social media, census data), unique patterns emerge.
The famous example: researchers re-identified Massachusetts Governor William Weld's medical records by linking "anonymized" hospital data with a public voter registration list using just ZIP code, birth date, and gender—quasi-identifiers that together were nearly unique.
Why Linkage Works
Attackers exploit overlapping quasi-identifiers across datasets. Your anonymized dataset might suppress names, but if it retains ZIP code, age, and occupation, an attacker can cross-reference these against:
- Public records databases
- Social media profiles
- Commercial data brokers
- Previously leaked datasets
Each additional dataset narrows the candidate pool until individuals become identifiable.
Defense Strategies
Reduce linkage surface area:
- Apply stronger generalization to quasi-identifiers (broader age ranges, coarser geographic regions)
- Use l-diversity and t-closeness to prevent attribute disclosure even after linkage
- Implement differential privacy to add calibrated noise that survives linkage attempts
Adversarial modeling:
- Assess what external datasets attackers could realistically access
- Measure re-identification risk under linkage scenarios, not just within your dataset
- Apply suppression or synthetic data generation for high-risk combinations
Contractual and technical controls:
- Restrict which fields are released together
- Use secure multi-party computation when cross-dataset analysis is necessary
- Monitor for attempted linkage attacks through access pattern analysis
Key Takeaway: Linkage attacks exploit the combination of multiple datasets to re-identify individuals; defenses require modeling attacker knowledge, strengthening quasi-identifier protection, and applying privacy techniques that resist cross-dataset correlation.