绿盟AI安全围栏 AI-GR(双语) - 自身安全产品 - 绿盟科技-巨人背后的专家

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

维护品牌与用户信任Protect brand and user trust

降低偏见、歧视、虚假信息等不当内容经AI应用输出的风险，为企业上线和持续运营AI应用提供支持。 Reduce the risk of AI applications producing biased, discriminatory, misleading, or otherwise inappropriate content, supporting both deployment and ongoing operations

02

保障业务连续性 Maintain service availability

针对算力DDoS等资源消耗型攻击提供防护，帮助维持服务可用性。分层快检、流式检测、安全代答及高可用部署，帮助兼顾安全处置、响应体验与服务稳定性。 Protection against resource consumption attacks, including distributed denial-of-service (DDoS) attacks targeting AI compute resources, helps maintain service availability Layered rapid inspection, streaming detection, safe alternative responses, and high-availability deployment balance protection, responsiveness, and service stability

03

保护数据与模型交互Protect data and model interactions

检测提示词攻击及个人隐私、商业机密泄露风险。通过语料安全检查，将部分风险识别前移至数据进入模型或知识库之前。 Detect prompt-based attacks and leakage risks involving personal information or business secrets Corpus security checks enable certain risks to be detected before data enters a model or knowledge base

04

支撑内容与数据合规Support content and data compliance

检测和处置违法违规内容，对敏感数据进行识别和脱敏，支持GB T 45654-2025等相关要求的落实。安全记录可为中国境内适用的生成式AI服务备案提供举证。 Detect and address prohibited content, identify and redact sensitive data, and support relevant requirements such as China s GB T 45654-2025 Security records can support filing procedures applicable to generative AI services in China

功能特性

Key Capabilities

01

安全报表与审计对接 Security reporting and audit integration

汇总防护资产、请求量、敏感数据拦截及攻击类型等信息。按需或按日、周、月生成HTML或PDF报表。分别保留审计与防护日志，并可转发至企业日志平台或安全运营中心（SOC）。 Report on protected assets, request volumes, sensitive data interceptions, and attack categories Generate HTML or PDF reports on demand or on a daily, weekly, or monthly schedule Retain audit and protection logs separately, with forwarding to enterprise logging platforms or a security operations center (SOC)
02

事件告警与追溯 Incident alerts and investigation

将同类风险聚合为事件，标注等级与研判标签，支持批量研判和处置。还原完整对话上下文，并通过邮件、短信、企业微信或钉钉发送告警，支持定位问题和调整策略。 Group related risks into incidents with severity and investigation labels for batch review and response Reconstruct full conversation context and deliver alerts via email, SMS, WeCom, or DingTalk to support investigation and policy refinement
03

安全替代答复 Safe alternative responses

在交互被拦截后，支持模板代答或模型生成代答。模板便于保持一致口径并减少时延，模型代答则提升交互自然度。涉及隐私风险时，使用固定模板兜底。 Return a template-based or model-generated response after an interaction is blocked Templates maintain consistent wording with lower latency, while model-generated responses provide more natural interaction Privacy-related cases use a fixed response template
04

隐私与敏感数据保护 Privacy and sensitive data protection

结合实体识别、误报抑制、敏感分级与审计记录，识别和保护个人信息及业务敏感数据。支持阻断、观察、脱敏三种模式，可针对输入与输出分别配置。 Entity recognition, false-positive suppression, sensitivity classification, and audit records support the identification and protection of personal information and business-sensitive data Blocking, monitoring, and redaction modes can be configured separately for inputs and outputs

05

输入输出双向检测 Bidirectional input and output inspection

输入侧重点检测攻击及违规诱导，输出侧审查模型生成内容。可按资产和策略分别配置检测阈值、开关及处置动作。检测结果包含风险类别、判定原因和检测引擎明细。 Inspect inputs for attacks and attempts to elicit prohibited content, and review model outputs for content risks Configure inspection thresholds, enablement, and response actions separately by asset and policy Results include risk categories, explanations, and engine-level findings
06

模型与智能体资产管理 Model and agent asset management

统一管理模型服务、AI应用和智能体资产，支持DeepSeek、通义千问、腾讯混元、文心一言、华为盘古、讯飞星火等模型。资产可手工创建或从外部系统批量同步，并自动生成唯一标识与访问凭证。业务流量切换至围栏入口后纳入防护。 Centrally manage AI applications, agents, and model services including DeepSeek, Qwen, Tencent Hunyuan, ERNIE, Huawei Pangu, and iFLYTEK Spark Create assets manually or synchronize them in bulk from external systems, with automatically generated unique identifiers and access credentials Protection applies when application traffic is routed through the guardrail endpoint

产品优势

Product Strengths

01

认证与生态兼容Certification and ecosystem compatibility

产品已通过公安部第三研究所相关认证及华为鲲鹏、昇腾生态兼容性认证。绿盟科技入选数世咨询《安全优先的大模型研究报告》推荐供应商名单。

The product has been certified by the Third Research Institute of China's Ministry of Public Security and for compatibility with the Huawei Kunpeng and Ascend ecosystems.

Digital World Consulting lists NSFOCUS as a recommended vendor in its Security-Prioritized Foundation Models report.

02

灵活策略与运行效率Configurable policies and operational efficiency

支持输入输出策略分离、检测引擎权重调整和模板化策略编排。采用轻量模型架构，结合流式检测与安全代答，控制硬件需求和安全处理开销。

Support separate input and output policies, adjustable inspection-engine weights, and template-based policy orchestration.

A lightweight model architecture, streaming inspection, and safe alternative responses help manage hardware demands and inspection overhead.

03

私有化与容器化部署Private and containerized deployment

支持容器化交付、私有化部署和Kubernetes编排，满足数据在本地处理的部署需求。适配华为鲲鹏、昇腾及麒麟、欧拉等软硬件环境。

Containerized delivery, private deployment, and Kubernetes orchestration support local data processing requirements.

Supported infrastructure includes Huawei Kunpeng and Ascend, together with Kylin and Euler operating systems.

04

资产与策略关联Policies linked to managed assets

将模型和智能体纳入统一资产管理，按资产关联安全策略，并支持批量同步和访问凭证生成，便于扩大已接入业务的防护覆盖。

Centrally manage models and agents as assets with associated security policies, using bulk synchronization and generated access credentials to extend consistent protection across connected applications.

05

多引擎语义检测Multiple engines for semantic detection

分层流水线与多检测引擎协同，覆盖38类合规风险和20类攻击分类。结合词法、语义及上下文，检测提示词注入、越狱、混淆编码和多模态注入等风险。

A layered pipeline coordinates multiple engines across 38 compliance risk categories and 20 attack categories.

Lexical, semantic, and contextual analysis addresses prompt injection, jailbreaks, encoded obfuscation, and multimodal injection.

成功案例

Success Cases

新能源电源企业Renewable energy power equipment enterprise

某新能源电源领域头部企业于2026年启动AI安全围栏项目，面向集团内大模型与智能体应用。项目采用插件或SDK接入、集群高可用部署及定制化集成，统一管理智能体资产与策略，并支持分权运维、敏感数据检测脱敏和安全事件追溯。 A leading renewable energy power equipment enterprise launched an AI-GR project in 2026 for group-wide LLM and agent applications Integration through plugins or software development kits (SDKs), high-availability clusters, and tailored integrations support centralized agent asset and policy management, delegated administration, sensitive data detection and redaction, and incident traceability

国家级国际博览会Major international expo in China

某国家级国际博览会使用大模型提供多语种客服和展台信息问答。AI-GR对风险会话进行实时检测，支持高并发业务运行，并留存会话内容，满足事后审计与追溯需求。 A major international expo in China introduced LLM-based multilingual customer support and exhibitor information services AI-GR inspected risky conversations in real time, supported high-concurrency service traffic, and retained conversations for subsequent audit and investigation

股份制商业银行Joint-Stock commercial bank

某股份制商业银行在手机银行、营销分析和国际业务中部署AI助手及智能体。AI-GR用于拦截AI客服不合规营销话术、检测和处置身份证号及电话号码等敏感信息，并防范恶意算力消耗，支撑相关AI应用的安全与合规运行。 A joint-stock commercial bank uses AI assistants and agents in mobile banking, marketing analytics, and international business AI-GR blocks noncompliant marketing responses from AI customer service applications, detects and handles sensitive information such as identity numbers and telephone numbers, and addresses malicious resource consumption to support secure and compliant AI operations

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

维护品牌与用户信任Protect brand and user trust

02

保障业务连续性 Maintain service availability

03

保护数据与模型交互Protect data and model interactions

04

支撑内容与数据合规Support content and data compliance

功能特性

安全报表与审计对接 Security reporting and audit integration

事件告警与追溯 Incident alerts and investigation

安全替代答复 Safe alternative responses

隐私与敏感数据保护 Privacy and sensitive data protection

输入输出双向检测 Bidirectional input and output inspection

模型与智能体资产管理 Model and agent asset management

产品优势

01

认证与生态兼容Certification and ecosystem compatibility

02

灵活策略与运行效率Configurable policies and operational efficiency

03

私有化与容器化部署Private and containerized deployment

04

资产与策略关联Policies linked to managed assets

05

多引擎语义检测Multiple engines for semantic detection

成功案例

新能源电源企业Renewable energy power equipment enterprise

国家级国际博览会Major international expo in China

股份制商业银行Joint-Stock commercial bank

您的联系方式

购买热线

购买咨询:

提交项目需求