AI Frontend Clarification and Input Filtering for LLM Security

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Large language model (LLM)-based chatbots face challenges in accurately interpreting user queries, leading to inaccurate responses and are vulnerable to attacks from attackers accessing their internal networks.

Innovation Solution

An artificial intelligence frontend system with input and output filter elements that execute clarification and filtering procedures to refine user queries and protect against adversarial attacks, using proactive learning, anomaly detection, and semantic similarity models to enhance query clarity and security.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If an LLM-based chatbot directly processes user queries without clarification, then the system responds quickly, but the accuracy and relevance of responses deteriorate due to vague or ambiguous queries

Engineering Contradiction:
Improveresponse accuracyVSAvoidquery processing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary clarification actions before final query processing. When a query is detected as vague or ambiguous, the system proactively asks clarifying questions to refine the user's intent before executing the actual query against the LLM, ensuring higher response accuracy while managing time through efficient clarification strategies

Inventive Principle:
Principle #10Preliminary action

2Reliability

If an LLM-based chatbot operates without input filtering, then the system is simple and fast, but the attack surface enlarges allowing attackers to invade the model without accessing internal networks

Engineering Contradiction:
ImprovesecurityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system introduces an intermediary filtering layer between the user input and the LLM processing engine. This intermediate component analyzes and filters incoming queries to detect and block malicious inputs, adversarial examples, or harmful content before it reaches the core model, thereby enhancing security without requiring direct modification of the complex LLM architecture

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system segments the query processing into distinct functional stages: input reception, security filtering, clarification handling, and LLM processing. By separating the security filtering function from the core model, the system can enhance reliability through specialized filtering mechanisms while maintaining the complexity management of the original LLM system

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4629098A1Artificial intelligence frontend system, artificial intelligence frontend operation method, and computer-readable recording medium with stored program
Publication Date: 2025.10.08 WISTRON CORP
  • EP4629098A1 patent drawingFigure 1~2
  • EP4629098A1 patent drawingFigure 3~4
  • EP4629098A1 patent drawingFigure 5

AI summary

An artificial intelligence frontend system (100), an artificial intelligence frontend operation method, a computer-readable recording medium with a stored program, and a non-transitory computer program product, where the artificial intelligence frontend operation method includes: executing an external clarification procedure in response to receiving a query by an input frontend (101) to obtain an external clarification query; directing the external clarification query to an input filter element (102); and executing a filtering procedure by the input frontend (101) on the external clarification query via the input filter element (102) to filter the external clarification query and sending an external clarification query which is filtered to an external chatbot (202) via the input filter element (102).