Secure and privacy-preserving query processing for large-scale language models
The method of identifying and substituting sensitive entities in queries for LLMs addresses privacy concerns in MLaaS, allowing secure and accurate use of large-scale language models.
JP2026520279APending Publication Date: 2026-06-23NEC LAB EURO GMBH
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- NEC LAB EURO GMBH
- Filing Date
- 2023-08-16
- Publication Date
- 2026-06-23
Smart Images
Figure 2026520279000001_ABST
Abstract
The computer implementation method enables a Large-Scale Language Model (LLM) to operate in a secure and privacy-protected manner. This method includes the steps of identifying entities within a query, where the query is configured to be processed by the LLM; generating a modified query by substituting the identified entities with semantically similar entities; and submitting the modified query to the LLM. The present invention can be used in a variety of applications, including, but is not limited to, several anticipated use cases in customer service, cybersecurity, finance, healthcare, and medical services.
Need to check novelty before this filing date? Find Prior Art