An employer wants to track how bonus compensations have changed over time to identify employee outliers and correct earning disparities. This task must be performed without exposing the sensitive compensation data for any individual and must be reversible to identify the outlier. Which Cloud Data Loss Prevention API technique should you use to accomplish this?

The requirement is to analyze compensation data for outliers while ensuring the data is not exposed but remains reversible. Format-Preserving Encryption (FPE), implemented via CryptoReplaceFfxFpeConfig, meets these needs. FPE replaces sensitive data with a surrogate token that maintains the original data format (e.g., a number remains a number). This process is deterministic, meaning the same input always produces the same token, which allows for tracking and analysis. Crucially, it is a cryptographic method that is reversible with the appropriate key, allowing authorized users to identify the original data for the outlier.

A. Generalization reduces data granularity (e.g., bucketing ages) and is not reversible, preventing the identification of specific individuals.
B. Redaction completely removes the data, making any form of analysis or reversal impossible.
C. CryptoHashConfig uses a one-way hashing function. While it provides a token, it is not reversible, which is a key requirement.

1. Google Cloud Documentation

"De-identification of sensitive data". In the "Tokenization" section

it states: "Unlike with hashing

tokenization is reversible... Format-preserving encryption (FPE)... is one way to produce tokens."

2. Google Cloud Documentation

"Transformations reference"

CryptoReplaceFfxFpeConfig. The documentation describes this as a method for format-preserving encryption

which "encrypts the input data and returns a base64-encoded encryption of the data." This encryption is reversible.

3. Google Cloud Documentation

"Balancing utility and privacy in data de-identification". This document discusses various techniques. For FPE

it notes its ability to "maintain referential integrity" (critical for analysis) and that it is "reversible through re-identification with the original cryptographic key."