Redacting Personally Identifiable Information in ChatGPT Content

As Generative A.I. continues to find a place in enterprises for information knowledge sharing and processing, organizations want a way to process the content of data while redacting (hiding / eliminating) personally identifiable information (PII) for privacy reasons.

Most tools to date "block" PII processing or requires the organization to manually "sanitize" content before ingesting it into their GenAI system. Microsoft now has a tool (currently in beta/Preview) that organizations can apply for early access to clean-up sensitive PII information, while retaining the core data of the content for GenAI information processing!

Microsoft Native Document Support for AzureAI

The technology Microsoft has available is called "Native Document Support for AzureAI" that is documented at https://learn.microsoft.com/en-us/azure/ai-services/language-service/native-document-support/use-native-documents?tabs=pii

It currently allows you to process PDF, DOCX, and TXT files

Effectively, upon processing of documents, it can turn an initial document with PII:

Into a document that has the PII redacted:

Preprocess or Pipeline the Redaction

To process the files for redaction, they can be run through a scripted routine to "clean-up" files before they are uploaded to a ChatGPT environment, or an automated process (aka a Pipeline) can be created using Microsoft's new Fabric technology that can automate the redaction process.

Although the content is being redacted, there is likely still sensitive information within the content, and thus using a PRIVATE instance of ChatGPT deployed using Microsoft's AzureAI/ChatGPT will provide an organization an internal instance of ChatGPT the Microsoft maintains privacy and security of content being processed by its OpenAI platform.

I cover the deployment and operations of AzureAI/ChatGPT in several articles I've posted on LinkedIn including - https://www.dhirubhai.net/pulse/leveraging-chatgpt-against-internal-office-365-create-rand-morimoto

Wrap-up

Just an additional step to leverage the power and capabilities of ChatGPT, in a safe, security, and compliant manner.