Personal data anonymization with model verification

A system comprising:

• a memory comprising instructions;
• and
• one or more processors, wherein the instructions, when executed by the one or more processors, cause the system to perform operations comprising:
• generating training data that includes anonymized communications data created by anonymizing communications data for a plurality of users, and personally identifiable information (PII) extracted from the communications data;
• training, based on the training data including anonymized communications data, a machine-learning program to generate a PII detection model that extracts PII values from input data;
• extracting, using the PII detection model executing on the one or more processors, PII values from a testing set that includes test anonymized communications data;
• comparing a performance of the PII detection model to extract the PII values from the test anonymized communications data to a performance of a guessing model that randomly guesses PII values in the test anonymized communications data;
• responsive to the comparing indicating that the performance of the PII detection model corresponds to the performance of the guessing model, determining that anonymization of the communications data based on the PII detection model is valid;
• and
• causing detection of malicious communications to be performed using the communications data anonymized based on the PII detection model.

A non-transitory machine-readable storage medium including instructions that, when executed by one or more processors of a machine, cause the machine to perform operations comprising:

• generating training data that includes anonymized communications data created by anonymizing communications data for a plurality of users, and personally identifiable information (PII) extracted from the communications data;
• training, based on the training data including anonymized communications data, a machine-learning program to generate a PII detection model that extracts PII values from input data;
• extracting, using the PII detection model executing on the one or more processors, PII values from a testing set that includes test anonymized communications data;
• comparing a performance of the PII detection model to extract the PII values from the test anonymized communications data to a performance of a guessing model that randomly guesses PII values in the test anonymized communications data;
• responsive to the comparing indicating that the performance of the PII detection model corresponds to the performance of the guessing model, determining that anonymization of the communications data based on the PII detection model is valid;
• and
• causing detection of malicious communications to be performed using the communications data anonymized based on the PII detection model.