Dynamic Network Analysis with Format-Preserving LLM Data Encryption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large and complex digital networks face challenges in information exchange and resource sharing due to interconnected device complexity, and existing encryption methods hinder the use of large language models by altering information formats, making it difficult for them to draw accurate conclusions.
Innovation Solution
A method and system for encrypting information using a large language model that maintains referential integrity and format preservation by replacing words with keys, ensuring consistent and understandable encrypted data for the model, while controlling sensitive information access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If standard encryption is used to protect sensitive network information, then security and protection from outside sources is improved, but the encrypted information becomes difficult for large language models to use and draw accurate conclusions
Solution Approach 1:
The patent changes the parameter of encryption from standard encryption to format-preserving encryption. This modifies the encryption output to maintain the original format (e.g., IP addresses remain as IP address format, URLs remain as URL format), allowing large language models to still process and understand the encrypted information while maintaining security protection.
Solution Approach 2:
The patent introduces a word-key pair database as an intermediary mechanism. This database maps encrypted words to their original forms, allowing the system to encrypt information before providing it to large language models while maintaining the ability to decode and understand the information when needed, thus bridging the gap between security and usability.
2Object-affected harmful factors
If encryption replaces words with keys to protect information, then sensitivity and security are improved, but referential integrity and consistent context are lost
Solution Approach 1:
The patent applies format-preserving encryption that maintains the structural parameters of the original text. Encrypted words retain their original format, length characteristics, and contextual relationships, ensuring that referential integrity is preserved while still providing security protection.
Solution Approach 2:
The patent creates a mapping copy between encrypted words and original words through the word-key pair database. This copy mechanism ensures that all references to a particular word are consistently replaced with the corresponding key, maintaining referential integrity across the entire document while preserving security.
3Productivity
If all network information is provided to the large language model for analysis, then comprehensive network analysis capability is improved, but information security and control over sensitive data is worsened
Solution Approach 1:
The patent segments the network information into two categories: information that needs to remain encrypted for security and information that can be processed by the large language model. The format-preserving encryption allows the system to provide encrypted segments to the model while maintaining security, and only decrypt necessary portions when needed for analysis.
Solution Approach 2:
The word-key pair database serves as an intermediary that allows selective decryption. The system can encrypt sensitive information before providing it to the large language model, but can selectively decrypt only the portions that need analysis, thus maintaining both security and comprehensive analysis capability.
Data Source
AI summary
A system for determining a natural language output regarding a digital network using a large language model (“LLM”) can include a computer processor configured to receive a desired output dependent upon information associated with the digital network; a prompt module configured to determine a query prompt to the LLM requesting the LLM to generate a query dependent upon the information and the desired output; and a graph database management system configured to determine, dependent upon a graph database representative of at least a portion of the digital network, a response to the query as received from the LLM, wherein the prompt module is also configured to determine an output prompt to the LLM requesting the LLM to generate the natural language output dependent upon the response to the query, and wherein the LLM generates the natural language output as requested in the output prompt.