Course contentsShow
← Security
Lesson 2908 of 3,05254. Privacy EngineeringFree lesson

Data Masking and Tokenization

Learn techniques to replace sensitive data with realistic but false values or reversible tokens.

Data Masking and Tokenization

What you'll learn: How to replace sensitive data with realistic but false values (masking) or reversible tokens to protect privacy while preserving utility.

After applying anonymization techniques like k-anonymity, you may still need operational ways to work with sensitive data safely. Data masking and tokenization provide practical methods to obscure real values during development, testing, analytics, or third-party sharing.

Data Masking

Data masking replaces original sensitive values with structurally similar but fictitious data. Think of it as replacing real Social Security Numbers with generated ones that look valid but don't belong to anyone.

Common techniques:

  • Substitution: Replace names with random names from a dictionary
  • Shuffling: Randomize values within a column across rows
  • Character masking: Replace digits with X's (e.g., "XXX-XX-1234")
  • Variance: Add/subtract random amounts to numeric values
  • Nulling: Replace with null/empty values

Masking is typically irreversible and best suited for non-production environments where analysts need realistic data patterns without actual sensitive information.

Tokenization

Tokenization replaces sensitive data with randomly generated tokens stored in a secure vault. Unlike masking, this is reversible: the original value can be retrieved using the token and vault access.

For example: credit card number 4532-1234-5678-9010 becomes token TKN-A7D9E2F1. The mapping lives in a highly secured token vault. Systems process tokens safely, only detokenizing when absolutely necessary (e.g., payment processing).

Benefits:

  • Reduces PCI DSS scope by keeping card data out of application databases
  • Enables secure data sharing across systems
  • Format-preserving tokenization maintains data structure (e.g., 16-digit tokens for 16-digit cards)

Key Takeaway: Data masking creates irreversible fake-but-realistic data for safe testing/analytics, while tokenization provides reversible replacement with vault-protected mapping—choose based on whether you need to recover original values.