Prompt Processing Units for Pre-Submission Data Policy Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data security approaches are ineffective in preventing sensitive data leakage and corporate policy violations due to the lack of understanding and natural-language native techniques in generative AI, rendering previous methods incapable of applying effective controls on data before processing by external entities.
Innovation Solution
The implementation of prompt processing units (PPUs) to characterize and distill key features from prompts, enabling systematic data control techniques that prevent sensitive data leakage and policy violations by applying targeted controls before sending prompts to external language models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional data security approaches are used, then data protection is maintained through conventional methods, but these approaches are ineffective against generative AI systems that lack understanding and natural-language native techniques
Solution Approach 1:
The patent introduces prompt processing units (PPUs) as intermediary components between the enterprise and external generative AI systems. These PPUs translate natural language prompts into structured formats that enable systematic data control techniques, acting as a mediator that bridges the gap between traditional security approaches and AI systems lacking native understanding capabilities.
Solution Approach 2:
The system performs preliminary characterization and distillation of key features from prompts before they are processed by external language models. By analyzing prompts in advance and identifying sensitive data patterns, the system can prevent unauthorized input and potential data leakage before the generative AI system processes the information.
2Reliability
If comprehensive data control policies are applied to all prompts, then sensitive data leakage is prevented, but processing delays and resource consumption increase
Solution Approach 1:
The patent implements selective data control by characterizing prompts and applying controls only where necessary. Instead of uniformly processing all prompts through comprehensive checks, the system identifies specific prompts containing sensitive data patterns and applies targeted controls only to those cases, thereby reducing overall processing delays while maintaining effective prevention of data leakage.
3Measurement precision
If detailed feature analysis of prompts is performed, then accuracy in detecting policy violations improves, but computational resources and processing time increase
Solution Approach 1:
The patent transforms prompts from unstructured natural language into structured characterized formats with extracted key features. This parameter transformation enables more efficient processing by converting complex linguistic analysis into structured data comparison, improving detection accuracy while reducing the computational resources required for subsequent policy violation detection.
Data Source
AI summary
In one implementation, a device may identify features of a prompt to be sent to a language model that are indicative of a type of data that the prompt would cause the language model to access. The device may determine whether data control policies apply for the prompt based on the type of data that the prompt would cause the language model to access. The device may determine, based on the features, whether processing of the prompt by the language model violates an applicable data control policy. The device may prevent the language model from processing the prompt when the processing of the prompt violates the applicable data control policy.