Selective anonymization for user data

A system for processing selectively anonymized user data, comprising:

• a memory;
• and
• at least one processor coupled to the memory and configured to:
• receive user data, wherein the user data comprises first data and a first prompt, and the first prompt indicates how to process the first data;
• receive an anonymization template, wherein the anonymization template specifies a profile to be anonymized in the user data and a tool for anonymizing the profile and wherein the profile specifies a type of confidential information within the user data to be anonymized;
• create, based on the anonymization template, anonymized user data by anonymizing the profile in the user data using the tool specified in the anonymization template, wherein the at least one processor is further configured to:
• determine the first data corresponds to the type of confidential information;
• and
• based on the determination, generate the anonymized user data by modifying the first data using a tag structure corresponding to the tool, wherein the first data is replaced by a character format specified by the tag structure;
• input the anonymized user data and the first prompt to a first language model, wherein the first language model evaluates the character format that replaced the first data;
• receive an anonymized response, wherein the anonymized response is a result of the first language model processing the anonymized user data and wherein the anonymized response maintains the tag structure;
• create a de-anonymized response by replacing the tag structure in the anonymized response with the first data;
• and
• output the de-anonymized response.

A non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one computing device, cause the at least one computing device to perform operations for processing selectively anonymized user data, the operations comprising:

• receiving user data, wherein the user data comprises first data and a first prompt, and the first prompt indicates how to process the first data;
• receiving an anonymization template, wherein the anonymization template specifies a profile to be anonymized in the user data and a tool for anonymizing the profile and wherein the profile specifies a type of confidential information within the user data to be anonymized;
• creating, based on the anonymization template, anonymized user data by anonymizing the profile in the user data using the tool specified in the anonymization template, wherein the creating further comprises:
• determining the first data corresponds to the type of confidential information;
• and
• based on the determining, generating the anonymized user data by modifying the first data using a tag structure corresponding to the tool, wherein the first data is replaced by a character format specified by the tag structure;
• inputting the anonymized user data and the first prompt to a first language model, wherein the first language model evaluates a character format that replaced the first data;
• receiving an anonymized response, wherein the anonymized response is a result of the first language model processing the anonymized user data and wherein the anonymized response maintains the tag structure;
• creating a de-anonymized response by replacing the tag structure in the anonymized response with the first data;
• and
• outputting the de-anonymized response.