LLM Sensitive Data Detection via Intermediary Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in enabling confidential computing operations for Large Language Models (LLMs) while maintaining the secrecy of both the algorithms and the sensitive data, particularly in healthcare where data protection regulations like HIPAA apply.
Innovation Solution
The system employs identifier determiner models and query sanitization models to detect and filter out sensitive information, ensuring that LLMs operate within a zero-trust environment without exposing confidential data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If LLMs are deployed to process sensitive healthcare data, then valuable insights and correlations can be identified, but confidential data may be exposed or exfiltrated
Solution Approach 1:
The patent introduces an identifier determiner model as an intermediary component between the LLM and the sensitive data. This model detects and identifies confidential information in the data inputs and outputs, acting as a mediator that prevents direct exposure of sensitive data while allowing the LLM to process it for valuable insights
Solution Approach 2:
The system performs preliminary identification of sensitive data elements before they are processed by the LLM. The identifier determiner model pre-screens inputs and pre-analyzes outputs to detect potential confidential information, enabling preventive measures to be taken before data exfiltration can occur
2Productivity
If proprietary algorithms are shared with data stewards for processing, then data can be processed efficiently, but algorithm secrecy and competitive advantage are compromised
Solution Approach 1:
The identifier determiner model serves as an intermediary that enables the data steward to understand and process the algorithm's outputs without actually seeing or accessing the proprietary algorithm itself. The model translates the algorithm's findings into identifiable sensitive data elements that can be processed further without exposing the underlying proprietary logic
3Measurement precision
If large datasets are transferred to algorithm developers for processing, then accurate analysis can be performed, but transfer time and bandwidth consumption increase significantly
Solution Approach 1:
The patent extracts the critical function of sensitive data identification from the data transfer process itself. By deploying the identifier determiner model at the data steward's location, the system extracts only the necessary analytical capability without requiring transfer of the entire large dataset, thereby maintaining analysis accuracy while dramatically reducing transfer time and bandwidth requirements
Data Source
AI summary
Systems and methods for the generation and usage of an identifier determiner model is provided. The identifier determiner model is generated in a sequestered computing node by receiving an untrained foundational model and a data set. The data set is bifurcated into a raw set and a de-identified set. The untrained foundational model is then tuned using the de-identified set to generate a sanitized model and the raw set to generate a raw model. Queries are presented to the raw model and the sanitized model to generate outputs. The identifier determiner machine learning model is generated by using the outputs to classify information as either sensitive or non-sensitive. The system may then receive a new foundational model. The identifier determiner machine learning model may be applied to outputs of this new foundational model to filter out sensitive information, either through redaction, or preventing them from being asked.