DEFINITION
Anonymization
Irreversibly transforming personal data so that no person can be identified, even with additional information.
In depth
Anonymization is the irreversible removal of identifying information from a dataset. Properly anonymized data is no longer personal data and falls outside scope of most privacy regimes (GDPR Recital 26). Anonymization is harder than it sounds: re-identification attacks combine quasi-identifiers (birth date, ZIP code, sex) and external datasets to recover identities from data that looked anonymous. Robust anonymization typically involves k-anonymity, l-diversity, differential privacy, or other formal techniques rather than simply 'removing names'.
Examples
- Aggregated statistics published with sufficiently large groupings that no individual can be re-identified.
- A dataset stripped of direct identifiers and processed through a differential-privacy mechanism before release.
- Using k-anonymity to ensure each record is indistinguishable from at least k-1 others.
How Locke handles anonymization
Locke prefers pseudonymization over anonymization for most prompt-protection use cases, because the user usually wants the AI to reason about a specific case — just not see the underlying identifiers. When full removal is appropriate, Locke's redaction mode produces output that is closer to anonymized than pseudonymized.
See the productRelated terms
- Pseudonymization
Replacing direct identifiers with stable pseudonyms so the data can no longer identify a person without additional information held separately.
- Redaction
Removing or blacking-out sensitive content so it is no longer present in the document or prompt.
- Personally Identifiable Information (PII)
Any information that can identify a specific person, either directly (name, SSN) or in combination with other data (zip code + birthdate).
