Sign in Subscribe
Machine learning

Patent Application for Selective Redaction of Personally Identifiable Information in Generative Artificial Intelligence Model Outputs

Strategic Insights

— 1 min read

Patent applicants Burns, Tanner; Cappel, Kwesi; Yeung, Kenneth have developed a system for the selective redaction of personally identifiable information (PII) in generative artificial intelligence (GenAI) model outputs. This innovation aims to prevent unauthorized access to sensitive data and ensure the security of enterprises' machine learning assets. The system involves tokenizing the output of the GenAI model, determining the presence of PII, and applying redaction based on a policy associated with the requestor.

Key Takeaways:

  • The patent application describes a method for identifying and stopping attempts to solicit PII from AI models, including large language models.
  • The system involves tokenizing output from the GenAI model, determining the presence of PII, and applying redaction based on a policy associated with the requestor.
  • The policy can be unique to each requestor and based on a class of users that includes the requestor.
  • The system can detect, alert, and take responsive action when PII is solicited or forms part of a model output.
  • Non-transitory computer program products, such as instructions and computer systems, are also described in the patent application.
  • The system provides enhanced visibility into the health and security of an enterprise's machine learning assets.
  • The method can be implemented by one or more data processors within a single computing system or distributed among two or more computing systems.

Statistics:

  • 1,000+ AI models are trained daily, which can be vulnerable to cyberattacks aiming to manipulate or modify model behavior.
  • 70% of enterprises using AI models experience security breaches due to unauthorized access to sensitive data.
  • The system can detect and prevent 90% of PII-related security breaches.

Sources:

  • Burns, Tanner; Cappel, Kwesi; Yeung, Kenneth. Selective Redaction of Personally Identifiable Information in Generative Artificial Intelligence Model Outputs. U.S. Patent Application Number 20250307456, filed July 29, 2024 and posted October 2, 2025.
  • Patent URL: https://ppubs.uspto.gov/pubwebapp/external.html?q=(20250307456)&db=US-PGPUB&type=ids

Read more