Prompt Engineering Gatekeeping for Secure LLM Question Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current prompt engineering in generative AI systems lacks sufficient security measures, allowing unauthorized users to access sensitive information and posing a risk of inappropriate or unanswerable responses.
Innovation Solution
A prompt engineering system that includes an acquisition unit, detection unit, extraction unit, and generation unit to analyze question data, detect user levels, and generate prompts that modify or refuse to answer based on the question's meaning and user authorization, ensuring secure interactions with large-scale language models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If prompt engineering allows high flexibility in asking questions, then ease of operation is improved, but security deteriorates due to unauthorized access to sensitive information
Solution Approach 1:
The patent introduces an intermediary component that acts as a mediator between the user's flexible questioning and the AI system. This intermediary analyzes the question, determines the questioner's level, and decides whether to modify or refuse the question before it reaches the AI, thus maintaining both flexibility and security.
Solution Approach 2:
The system performs preliminary analysis of the question and questioner level before the AI processes it. By determining the questioner's authorization level in advance and pre-modifying or pre-refusing inappropriate questions, the system prevents security issues before they occur while maintaining operational flexibility.
2Ease of operation
If the system accepts all questions from any user, then ease of operation is improved, but harmful factors increase due to unauthorized access to sensitive information
Solution Approach 1:
The system applies preliminary anti-action by proactively identifying and neutralizing potential harmful questions before they can cause damage. The question analysis unit detects sensitive information requests, and the system refuses or modifies these questions in advance, preventing unauthorized access while maintaining open accessibility for legitimate queries.
3Productivity
If the system processes all questions without filtering, then productivity is improved, but reliability deteriorates due to inappropriate responses
Solution Approach 1:
The system extracts and separates the security analysis function from the main question processing flow. By taking out the question analysis and level determination as independent preprocessing steps, the system can quickly filter inappropriate questions before they enter the main AI processing pipeline, maintaining productivity while improving reliability through targeted filtering.
Data Source
AI summary
A prompt engineering computer that generates a prompt for input into a large-scale language model is provided. The prompt engineering computer acquires question data, detects a questioner level, extracts a first keyword from the question data; determines whether or not a sentence meaning of the question data is answerable based on the first keyword and the questioner level, and generates a prompt that modifies at least a part of the first keyword or a prompt that refuses to answer if the sentence meaning is unanswerable.