Data Anonymization vs Pseudonymization: GDPR-Compliant De-identification Techniques

De-identification Under GDPR

The GDPR distinguishes between anonymization (irreversible) and pseudonymization (reversible with additional information). This distinction has significant implications for how data can be used and shared.

Anonymization vs Pseudonymization

Anonymization Techniques

K-Anonymity

Ensure each record is indistinguishable from at least k-1 other records:

# Before (k=1, unique)
Age=29, ZIP=10001, Disease=Flu

# After (k=3, at least 3 identical quasi-identifiers)
Age=25-30, ZIP=100**, Disease=Flu

Differential Privacy

Add calibrated noise to query results:

# True count: 42
# With differential privacy (epsilon=1.0):
noisy_count = true_count + laplace_noise(sensitivity=1, epsilon=1.0)
# Result: ~42 (varies each query)

Data Masking

Replace sensitive values with realistic fakes:

# Original: John Smith, 555-0123, john@email.com
# Masked:   Jane Doe, 555-9876, jane@masked.com

Pseudonymization Techniques

Tokenization

Replace sensitive values with random tokens:

Original: SSN 123-45-6789
Token:    tok_a7f3e2b1c4d5

A secure token vault maps tokens to original values.

Encryption

Encrypt identifiers with a key stored separately:

pseudonym = encrypt(identifier, key)
# Key stored in separate, access-controlled system

Hashing with Salt

import hashlib
pseudonym = hashlib.sha256((salt + identifier).encode()).hexdigest()

Implementation Best Practices

• Choose technique based on use case and regulatory requirement
• Test for re-identification risk before sharing
• Protect pseudonymization keys with the same rigor as the data
• Document the de-identification methodology
• Regular re-assessment as new data sources emerge

Where SuperClue Fits

SuperClue does not scan your systems for personal data or recommend de-identification techniques. Use the techniques above, and use its DPDP framework scan to check what your public website shows about personal data handling.