绿盟AI安全一体机 AI-UTM（双语） - 自身安全产品

AI安全
安全产品与方案
行业解决方案
安全服务与运营
安全研究
合作伙伴
技术支持

风云卫AI赋能安全系列
清风卫AI自身安全系列

风云卫AI赋能安全系列

赋能安全产品

基础设施安全
数据安全
云计算安全
工业互联网安全
物联网安全
信息技术应用创新
全部产品
全部解决方案

基础设施安全

网络安全

安全管理

应用安全

终端安全

身份与访问管理

解决方案

客户价值

Customer Benefits

01

适配现有IT环境Fit existing IT environments

支持x86与ARM架构、IPv4与IPv6双栈，以及旁路和网关部署模式，通过标准化接口对接企业现有安全体系和多模型环境。 Support for x86 and ARM, IPv4 IPv6 dual stack, out-of-band monitoring, and gateway deployment, together with standardized interfaces, enables integration with existing security systems and environments using multiple models

02

降低AI应用风险Reduce AI application risk

将内容安全、应用防护、数据防泄漏与安全评估结合，降低敏感数据泄露、应用越权及不当内容输出等风险。 Combining content inspection, application protection, data loss prevention, and security assessment helps reduce exposure to sensitive data leakage, unauthorized application access, and inappropriate model outputs

03

保护算力与关键业务Protect compute resources and critical services

通过Token配额、访问控制、速率限制和多级优先级调度，限制异常资源消耗与Token耗尽攻击，为高优先级业务保留模型服务资源。 Token quotas, access controls, rate limits, and multi-tier priority scheduling help contain abnormal resource consumption and token exhaustion attacks, preserving model service capacity for higher-priority workloads

04

支撑合规运营Support compliance

通过内容审核、用户权限控制、日志追溯及策略可视化，支持企业落实《生成式人工智能服务管理暂行办法》及TC260-003《生成式人工智能服务安全基本要求》中的相关要求。 Content review, access permissions, traceable logs, and policy visibility help organizations address applicable requirements under China s Interim Measures for the Management of Generative AI Services and the TC260-003 technical guidance on basic security requirements for generative AI services. 审计记录为安全审计提供依据。 Audit records provide evidence for security reviews.

功能特性

Key Capabilities

01

内置大模型安全评估 Integrated LLM security assessment

内置AI-SCAN安全评估能力，通过API向目标模型提交不同类型的文本测试集。评估参照《生成式人工智能服务安全基本要求》，结合内容相似性、语义一致性等分析输出安全评分。 Integrated AI-SCAN assessment capabilities submit different types of text test datasets to target models through APIs. Assessments reference China s basic security requirements for generative AI services and use measures such as content similarity and semantic consistency to produce security scores.
02

敏感数据防泄漏 Sensitive data loss prevention

识别交互流量中的个人身份信息、商业机密等敏感内容。对命中信息进行实时掩码或替换，并对敏感文件实施拦截与隔离，降低知识库数据和个人隐私经AI交互泄露的风险。 Inspect interaction traffic for personally identifiable information (PII), business secrets, and other sensitive data. Real-time masking or replacement of detected data, together with blocking and isolation of sensitive files, helps reduce the risk of knowledge-base data and personal information leaking through AI interactions.
03

大模型应用安全防护 LLM application protection

检测和阻断提示词注入、越狱与混淆攻击，提供API访问控制及恶意扫描防护。在网关部署模式下，检测并阻断利用Ollama、LangChain、RAGFlow、AgentScope等组件漏洞的攻击。 Detect and block prompt injection, jailbreaks, and obfuscated attacks, with API access controls and protection against malicious scanning. In gateway deployment mode, detect and block attacks exploiting vulnerabilities in components such as Ollama, LangChain, RAGFlow, and AgentScope.
04

AI网关与访问身份管理 AI gateway and identity-based access control

通过虚拟API Key进行认证和授权，支持资源使用治理、成本统计与细粒度审计。reporting, and detailed auditing 全局Token每分钟（TPM）与请求每分钟（RPM）限制、多级优先级调度及Token配额，帮助控制过载与异常消耗。 Virtual API keys provide authentication and authorization, with resource usage governance, cost Global tokens-per-minute (TPM) and requests-per-minute (RPM) limits, multi-tier priority scheduling, and token quotas help control overload and abnormal consumption

05

内容安全检测与处置 Content inspection and response

结合词法、语义与上下文检测，覆盖违法违规、歧视偏见、隐私泄露、商业违法等50余类细分风险。支持本地关键词检测与安全围栏语义检测，对高风险内容实施阻断，并向用户提供安全替代答复。 Lexical, semantic, and contextual analysis covers more than 50 risk subcategories, including prohibited content, discrimination, privacy leakage, and unlawful business practices. Local keyword detection and AI guardrail semantic analysis can block high-risk content and return a safe alternative response.
06

大模型资产统一管理 Centralized LLM asset management

支持纳管DeepSeek、通义千问、腾讯混元、文心一言、华为盘古、讯飞星火等模型服务，以及通过产品完整代理的AI应用。按照资产名称、协议、URL、API Key等信息管理接入对象，并按资产配置和下发防护策略。 Manage model services including DeepSeek, Qwen, Tencent Hunyuan, ERNIE, Huawei Pangu, and iFLYTEK Spark, together with AI applications fully proxied through the product. Manage connected assets by name, protocol, URL, and API key, and configure and apply protection policies for each asset

产品优势

Product Strengths

01

兼顾并发与流式体验Designed for concurrent and streaming workloads

通过多检测引擎并发异步处理、服务端事件（SSE）流式内容断句和上下文缓存，减少安全检测对交互时延的影响。

Concurrent asynchronous inspection engines, sentence segmentation for server-sent event (SSE) streams, and context caching help reduce the latency introduced by security inspection.

02

开放集成与扩展Integration and extensibility

采用模块化架构，支持与绿盟TONE平台联动，并通过标准化接口接入外部系统。支持算力升级、模型替换和知识库扩展。

A modular architecture supports integration with the NSFOCUS TONE platform and external systems through standardized interfaces.

Compute resources can be upgraded, models replaced, and knowledge bases expanded.

03

结合模型推理与安全检测Model-Assisted threat detection

依托绿盟自研风云卫大模型与安全围栏，结合语义分析、检索增强、提示工程、行为建模和异常检测，分析复杂攻击及异常交互。

NSFOCUS's proprietary security model, NSFGPT (Fengyunwei), and AI guardrail capabilities combine semantic analysis, retrieval augmentation, prompt engineering, behavioral modeling, and anomaly detection to analyze complex attacks and suspicious interactions.

04

围绕AI资产配置防护Protection organized around AI assets

将已纳管模型服务与AI应用作为策略管理对象，使防护规则与具体资产关联，便于统一运营、审计和资源使用管理。

Protection policies are associated with managed model services and AI applications, bringing security operations, auditing, and resource usage management into a consistent asset-based model.

场景

Use Cases

AI交互风险监测与防护Monitor and protect AI interactions

通过旁路部署监测AI应用出口流量，识别敏感话题、敏感数据及攻击行为并留存审计记录。在流量经过网关的部署模式下，可按策略执行阻断和安全代答。 Out-of-band deployment monitors AI application egress traffic for sensitive topics, sensitive data, and attacks, retaining audit records When traffic is routed through the gateway, p

模型资源与API使用管理Govern model resources and API usage

通过Token限流、API Key管理及面向资源耗尽攻击的防护，控制异常调用，为高优先级应用分配算力与带宽资源。 Token rate limits, API key management, and defenses against resource exhaustion help control abusive calls and prioritize compute and bandwidth for higher-priority AI applications

分支与远程用户访问审计Audit branch and remote user access

对访问大模型应用的分支及远程用户进行认证与授权，记录、还原和分析交互内容。结合防火墙联动，可追溯网络地址转换（NAT）前的源IP。 Authenticate and authorize branch and remote users accessing LLM applications, then record, reconstruct, and analyze their AI interactions Firewall integration supports tracing the origi

风云卫AI赋能安全系列

赋能安全产品

清风卫AI自身安全系列

自身安全产品

解决方案

全部产品查看更多 >

全部解决方案查看更多 >

基础设施安全

网络安全

安全管理

应用安全

终端安全

身份与访问管理

解决方案

数据安全

数据安全产品

解决方案

云计算安全

云计算安全产品

解决方案

工业互联网安全

工业互联网安全产品

解决方案

物联网安全

物联网安全产品

解决方案

信息技术应用创新

信创类安全产品

全部产品查看更多 >

全部解决方案查看更多 >

政府查看更多 >

解决方案

运营商查看更多 >

解决方案

金融查看更多 >

解决方案

能源查看更多 >

解决方案

交通查看更多 >

解决方案

企业查看更多 >

解决方案

科教文卫查看更多 >

解决方案

产品概述

客户价值

01

适配现有IT环境Fit existing IT environments

02

降低AI应用风险Reduce AI application risk

03

保护算力与关键业务Protect compute resources and critical services

04

支撑合规运营Support compliance

功能特性

内置大模型安全评估 Integrated LLM security assessment

敏感数据防泄漏 Sensitive data loss prevention

大模型应用安全防护 LLM application protection

AI网关与访问身份管理 AI gateway and identity-based access control

内容安全检测与处置 Content inspection and response

大模型资产统一管理 Centralized LLM asset management

产品优势

01

兼顾并发与流式体验Designed for concurrent and streaming workloads

02

开放集成与扩展Integration and extensibility

03

结合模型推理与安全检测Model-Assisted threat detection

04

围绕AI资产配置防护Protection organized around AI assets

场景

AI交互风险监测与防护Monitor and protect AI interactions

模型资源与API使用管理Govern model resources and API usage

分支与远程用户访问审计Audit branch and remote user access

您的联系方式

购买热线

购买咨询:

提交项目需求

欢迎加入绿盟科技，成为我们的合作伙伴！

您的联系方式