LLM Caller Authentication Using Personalized User Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies for identifying and addressing the challenges of malicious third-party access to user accounts are not foolproof and are often exploited by malicious third parties, leading to unauthorized access and information theft.

Innovation Solution

Utilizing a personalized large language model (LLM) trained on user-specific data and a general-purpose LLM to generate tailored identification questions, which are answered and compared against user data to authenticate the caller's identity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If generic predetermined questions are used for authentication, then the system is simple to operate, but the authentication effectiveness deteriorates because the questions are easily found online and are not personalized to the user

Engineering Contradiction:

Improveease of operationVSAvoidauthentication effectiveness

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The patent applies local quality by transitioning from generic questions to personalized questions that are specifically tailored to each user based on their unique data, behaviors, and context. The system generates authentication questions that are locally relevant to the specific user being authenticated, making the questions difficult for attackers to predict while maintaining ease of use through automated question generation and administration.

Inventive Principle:

Principle #3Local quality

2Ease of operation

If a list of predetermined questions and answers is maintained, then the authentication process is simple, but the system becomes complex to maintain and update regularly to remain effective

Engineering Contradiction:

Improveease of operationVSAvoidsystem maintenance complexity

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The system implements self-service by automatically generating and updating authentication questions based on user data, behaviors, and context without requiring manual intervention to maintain the question list. The automated question generation system continuously adapts to changing user patterns and eliminates the need for human administrators to manually update questions, thereby reducing maintenance complexity while maintaining high authentication effectiveness.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent applies parameter changes by dynamically adjusting authentication questions based on changing user behaviors, data patterns, and contextual information. Instead of using a static list of predetermined questions, the system continuously evolves the questions based on real-time analysis of user activity, making the authentication mechanism adaptable to new threats and user patterns without manual reconfiguration.

Inventive Principle:

Principle #35Parameter changes

3Ease of manufacture

If publicly available information is used for verification questions, then the questions are easy to generate, but the authentication security deteriorates because malicious third parties can easily find and exploit this information

Engineering Contradiction:

Improveease of question generationVSAvoidsecurity vulnerability to information theft

Core Design Contradiction:

Ease of manufactureVSObject-affected harmful factors

Solution Approach 1:

The system extracts authentication questions away from publicly available information sources and generates them from private, proprietary data about the user's behaviors, patterns, and context. By taking out the question generation process from public data and grounding it in private user-specific information, the system eliminates the vulnerability where attackers can easily find verification answers online while maintaining ease of automated question generation.

Inventive Principle:

Principle #2Taking out (Extraction)

4Reliability

If automated voice assistants ask identification questions to callers, then the system can prevent some malicious access, but the authentication remains not foolproof and can be deceived

Engineering Contradiction:

Improveauthentication capabilityVSAvoiddeception by malicious third parties

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent applies mechanics substitution by replacing traditional mechanical authentication methods (fixed question lists, simple secrets) with an intelligent system that uses AI-generated questions based on user behavior patterns, contextual analysis, and proprietary data. This substitution of the authentication mechanism from static to dynamic, intelligent question generation significantly improves reliability while making deception by malicious third parties much more difficult.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20250371178A1Authentication and identification of third parties using general and personalized large language models

Publication Date: 2025.12.04 GEN DIGITAL INC

AI summary

A third-party attempting to call a user is identified and authenticated by collecting multi-dimensional user data and user activity into a personalized large language model (LLM) residing on a user device. A general-purpose LLM generates third-party authentication questions based on an initial intent of the caller and the questions are provided to the personalized LLM with the answers from the personalized LLM ranked based on relevancy and accuracy. One or more of the highest ranked questions are prompted to the caller and the answers provided by the caller are compared to the answers provided by the personalized LLM to determine if the caller is a trusted contact or a likely malicious third-party.