Screen Content Desensitization for Sensitive AI Sharing Sessions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content sharing methods between terminals and large models are vulnerable to sensitive content leakage during screen sharing sessions.
Innovation Solution
A content sharing method that performs content desensitization processing on sensitive content, replacing it with its type, and type labeling, ensuring only the type of sensitive content is shared with the large model, thereby preventing the model from learning or outputting the actual sensitive content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If screen content is shared with the large model, then the model can access and process the content, but sensitive content may be leaked
Solution Approach 1:
The patent extracts sensitive content from the shared screen content through content recognition and identification. The sensitive information is separated from the overall content stream, allowing the large model to process only the non-sensitive portions while the sensitive data is excluded, thus preventing leakage while maintaining content accessibility.
Solution Approach 2:
The patent introduces an intermediary processing layer between the screen content and the large model. This intermediary performs content desensitization processing, including masking, blurring, or replacing sensitive content with placeholder symbols, thereby mediating between content sharing and security protection.
2Object-affected harmful factors
If content desensitization processing is performed on sensitive content, then sensitive content leakage is prevented, but processing complexity increases
Solution Approach 1:
The patent performs content recognition and sensitive content identification in advance before the actual sharing process. By pre-processing the content to identify and mark sensitive areas, the system reduces the complexity of real-time processing during sharing, as the desensitization targets are already determined.
Solution Approach 2:
The patent changes the state or form of sensitive content through various desensitization parameters such as masking degree, blurring intensity, or placeholder selection. These parameter adjustments allow flexible control over the desensitization process, managing complexity while maintaining security.
3Object-affected harmful factors
If type labeling processing is performed on sensitive content positions, then the large model can understand content structure without accessing sensitive data, but processing time increases
Solution Approach 1:
The patent performs type labeling processing as part of the preliminary content preparation stage. By pre-adding position information and content type labels to the desensitized content, the system avoids the need for complex real-time analysis during the sharing process, thus reducing processing time while maintaining the large model's ability to understand content structure.
Data Source
AI summary
A content sharing method includes: obtaining a first multimedia content to be shared in a terminal during a session between the terminal and a large model; performing a content desensitization processing on a sensitive content in the first multimedia content, and obtaining a second multimedia content by performing a type labeling processing on a position of the sensitive content in the first multimedia content, in which a type labeled by the type label processing is a type of the sensitive content; and sharing with the large model the second multimedia content for the session.