Configure sensitive data handling for generative AI
Set up and configure how personally identifiable information and other sensitive data is removed from generative AI prompts.
Antes de Iniciar
You can create a new Sensitive Data Regex record with a pattern or you can modify an existing record. The Generative AI category will apply the pattern to your generative AI prompts.
Role required: admin
Por Que e Quando Desempenhar Esta Tarefa
Personally identifiable information and other sensitive data can be masked so that it does not appear in generative AI prompts. Placeholder text is sent with the prompt instead, and that placeholder text is replaced with the original text after the response has been received. This two-way masking ensures your users see the correct values, but the LLM is not exposed to any sensitive information.
- The sensitive data detection toggle in the Conversational Interfaces Settings is unrelated to the Generative AI Controller, which uses sensitive data handling capabilities separately from this utility.
- The prompt, response, edited_response and additional_data fields in a conversational topic are masked during data extraction.
- The prompt is masked before sending it to the large language model (LLM) by using the sensitive data handling plugin. Only regular expressions (regexes) in the Generative AI category are used to mask the prompt.