Topsec LLM Security Gateway System

Large Model Security | Prompt Injection Attack Protection | Content Security & Compliance | TopLMG

Product Overview

Topsec LLM Security Gateway System adopts the multi-level security protection concept for large models and a built-in pre-trained security detection model. It provides core technical capabilities such as AI recognition, large model session parsing, application and API protection, prompt injection protection, and value-oriented content filtering for large model application systems. It creates a five-layer in-depth defense system covering the infrastructure, service, user, content and supervision layers of large models, and innovatively builds an intelligent defense closed loop of "Identify–Protect–Detect–Respond–Recover", delivering all-round protection from the network layer to the application layer for enterprise-level large model services.

Activity & Announcement

Hot Products Intelligent Computing Cloud Platform
New Products Ask Large Model
New Products LLM Security Gateway
Typical cases CNCC security construction project

Advantages

Prompt Injection Attack Protection

Equipped with a built-in pre-trained model for detecting prompt injection attacks, it provides detection and protection against such attacks including model hijacking, role-playing, developer mode, DAN jailbreaking, adversarial suffix attacks, random noise and weak semantics.

Value-Oriented Content Filtering

Based on the massive collected value-oriented data, it completes the training and construction of a value-oriented model. After parsing user input prompts or restoring response streams, the value-oriented model is activated to conduct content detection. It prevents the model from outputting content that violates national laws and regulations, involves politically sensitive elements or contains incorrect values.

Sensitive Data Leakage Prevention

It supports the sensitive data leakage prevention function, enabling real-time filtering of sensitive data and resisting malicious users from stealing sensitive information such as ID numbers, phone numbers and key business data in the large model database through illegal means. The system has a rich built-in library of sensitive data types and also supports users to flexibly customize various sensitive data types according to their own business needs.

Supply Chain Vulnerability Protection

As a core component of intelligent system protection, it builds a professional and systematic defense system against vulnerability attacks. Its built-in rule base for large model-related components and vulnerability features is deeply optimized, covering a variety of classic and new attack types such as directory traversal, SQL injection, XSS (Cross-Site Scripting) and SSRF (Server-Side Request Forgery). It can accurately identify potential risks in links such as code logic, protocol parsing and permission control for large model components and frameworks.

API Security Protection

It builds a full-process and intelligent API protection system to provide three-dimensional security guarantees for the external service interfaces of large models. At the API asset governance level, it supports API asset import and automatic discovery capabilities. In terms of risk prevention and control, the security gateway provides API compliance verification and access control functions; for API abuse scenarios, the system supports fine-grained access rate limiting policies.

2025

June 2025

Awarded the first domestic certificate for Large Model Security Protection Fence Products (Enhanced Level) in China.

Applications

Customer-Owned Large Model Scenario

Deployed in front of the customer's large model via a transparent proxy mode, it enables ultra-simple access through non-intrusive deployment without any configuration modifications to the client and server sides. Realizing plug-and-play with zero transformation for business parties and zero perception for users, it inspects all business traffic to and from the large model, effectively identifies and blocks irregularities such as prompt injection attacks, and filters output content to avoid compliance risks caused by the output of non-compliant and sensitive words.

Value to Customers

Compliance Supervision

It detects and filters input and output content to avoid compliance risks arising from the output of content that contravenes mainstream values or politically sensitive terms.

Data Security

It defends against malicious users stealing sensitive information from the large model database through illegal means, such as ID numbers, phone numbers, key business information and other sensitive data.

Interaction Security

It prevents the circumvention of the large model's risk control module via prompt injection attacks, which could induce the model to output harmful, biased, discriminatory or unethical content.

API Security

It conducts resource management and control of API assets through functions including compliance verification, access control and access rate limiting, minimizing the scope of attack impact and preventing resource abuse.

Relevant Recommendations

Product Recommendation

Special Recommendation

News Recommendation