AI Model Access Control via Data Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models lack a mechanism to control access to input data based on access rights, leading to the potential leakage of confidential or sensitive information.
Innovation Solution
A method that refines information available for AI models using access rights or access control policies to obtain filtered data, which includes learning information and request information, ensuring that only authorized data is processed and output by the AI model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If access control mechanisms are implemented for AI models, then information security is improved, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by establishing access control policies and refining information before it is input to the AI model. The controller pre-processes information to extract only authorized content based on access rights, so that when the AI model receives the refined information, the access control framework is already in place, preventing unauthorized data from being processed in the first place.
Solution Approach 2:
The patent introduces a controller as an intermediary component between the information source and the AI model. This controller acts as a mediator that receives original information, applies access control policies, and outputs refined information to the AI model. This intermediary layer enforces security rules without requiring modifications to the AI model itself, thus improving security while maintaining relative system simplicity.
2Productivity
If all available information is provided to AI models, then model performance is improved, but risk of confidential information leakage increases
Solution Approach 1:
The patent applies local quality by differentiating the quality and accessibility of different portions of information. Instead of treating all information uniformly, the system applies access control policies selectively to specific information sources and types. The controller refines information by preserving authorized content while filtering unauthorized content, giving different 'quality' levels to different data portions based on their access rights.
Solution Approach 2:
The patent changes the parameter of information availability by transforming the complete information set into a refined information set. The controller modifies the information parameters by applying access control policies, changing which data points are accessible to the AI model. This parameter change ensures that only information within authorized bounds is provided, reducing leakage risk while maintaining model performance with the available authorized data.
3Reliability
If access control filtering is applied to information, then data security is improved, but information availability for processing decreases
Solution Approach 1:
The patent applies partial action by filtering only the necessary portions of information that require access control. The controller does not blanket-filter all information but selectively applies access control policies to specific information sources. This partial filtering approach maintains data security for sensitive information while preserving availability of authorized information, minimizing the loss of useful data.
Solution Approach 2:
The system implements feedback mechanisms where access control policies are applied and the refined information is provided to the AI model. The controller continuously monitors and adjusts the filtering process based on the access rights and policies, ensuring that security requirements are met while maximizing the availability of authorized information for processing.
Data Source
AI summary
A method and a device for performing access control based on an access right for an artificial intelligence (AI) model are disclosed. A method for controlling access to at least one AI model according to an aspect of the present disclosure includes refining information available for the at least one AI model based on at least one of access rights or access control policies to obtain filtered data for the at least one AI model; and obtaining output data of the at least one AI model based on the filtered data, and the filtered data may include at least one of learning information for the at least one AI model or request information for the at least one AI model.