AI Prompt Sanitization for Sensitive Data and Intent Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing generative artificial intelligence (GAI) systems face challenges in handling sensitive data, as they may inadvertently retain or misuse confidential information from user prompts, posing security risks and loss of intent.
Innovation Solution
A sanitization system that analyzes user prompts to determine the context and intent, selectively modifying or removing sensitive data before forwarding it to GAI systems, using domain-specific models to maintain the original intent while reducing data exposure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensitive data is removed or modified from user prompts, then data security is improved, but the original intent of the message may be lost
Solution Approach 1:
The system segments the user prompt into individual phrases and analyzes each phrase separately using domain-specific models. This allows selective modification of only those phrases containing sensitive data while preserving the intent-critical portions of the message, thereby resolving the contradiction between security and intent preservation.
Solution Approach 2:
The system applies different processing treatments to different parts of the prompt based on their sensitivity and importance characteristics. Phrases are evaluated individually, and only specific local portions containing sensitive data are modified, while other portions remain unchanged to maintain the original message intent.
2Loss of information
If domain-specific models are used to analyze each phrase, then intent preservation is improved, but system complexity increases
Solution Approach 1:
The system employs a unified multi-functional architecture where domain-specific models serve multiple purposes: they identify sensitive data, assess phrase importance, and generate modifications. This universal approach handles diverse domains (medical, legal, financial, etc.) through a single system framework, reducing overall complexity despite the sophisticated analysis required.
3Loss of information
If all phrases are provided to the external system, then completeness of information is improved, but risk of data leakage increases
Solution Approach 1:
The system performs preliminary analysis and modification of the user prompt before transmitting it to the external generative AI system. By proactively identifying and redacting sensitive information in advance, the system ensures that the transmitted message is both as complete as possible for maintaining intent and sanitized to prevent data leakage.
Solution Approach 2:
The system acts as an intermediary layer between the user and the external GAI system. It receives the original prompt, processes it through domain-specific models to identify and modify sensitive phrases, and then transmits the sanitized version to the external system, thereby mediating the information flow to balance completeness and security.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for sanitizing artificial intelligence prompts. One of the methods includes receiving a message a) for an external system and b) that comprises two or more phrases; for at least one phrase from the two or more phrases: determining a context of the phrase in the message; determining, using the context, whether modification of the phrase will likely maintain an intent of the message; determining whether to permit unedited transmission of the message to the external system using a result of at least one of one or more determinations whether modification of the phrase will likely maintain the intent of the message; and performing one or more actions using a result of the determination whether to permit unedited transmission of the message to the external system.