Takes over real AI exits.接管真实 AI 出口。
Deploy at every AI exit point. Observe browser, API, copilot, and agent traffic before it leaves your boundary.在每一个 AI 出口部署，于数据离域前观察浏览器、接口、协同助手与智能体流量。
Mode IO AI Governance PlatformMode IO · AI 安全治理
See Every AI Action. Prove Every Decision.AI 安全治理与隐私保护
Enforcement across gateway, cloud APIs, and plugins for AI security, privacy, and cross-border compliance. Risk-appropriate enforcement at request time, with explainable audit evidence on every request.覆盖网关、云端 API 与插件的 AI 安全、隐私与跨境合规。请求时按风险分级处置，每次请求留存可解释审计证据。
Product demo产品演示
Preview预览
Governance Overview治理总览
Explore risk posture, ASI distribution, and audit trails in our interactive console.查看风险态势、ASI 分布与审计留痕。
Simulated模拟
Open Governance Overview打开治理总览演示
Risks
12.8K
Success
99.7%
Agents
184
Tokens
51M
ASI 01–10
Real-Time AI Security Audit实时 AI 安全审查
Simulated模拟
Why this architecture为什么这样设计
Capture earlier. Decide faster. Prove later.更早捕获、更快判断、事后可证。
Capture traffic, review with small models, enforce policy, and retain audit evidence. Layers onto your existing stack.捕获流量、小模型审查、策略处置与证据留存，叠加于现有技术栈。
Turns risk into enforceable actions.把风险转化为可执行动作。
Small models return verdicts mapped to risk-appropriate actions. P0–P3 configurable per context.小模型输出判断，映射为按风险等级匹配的动作；P0–P3 可按场景配置。
Policy version, judgment basis, audit evidence.策略版本、判断依据、审计证据。
Every verdict records policy version, judgment basis, and audit evidence—with explainable rationale so reviewers see why, not just what.每次判断留存策略版本、判断依据与审计证据；处置逻辑可解释，复核与审计能还原「为何如此」。
Traffic architecture流量架构
Every employee and agent routes through governance before transmission to external AI.每一位员工与智能体，在向外部 AI 传输前先经治理层审查。
Traffic sources流量来源
Employees & Agents员工与智能体
Governance layer治理层
Gateway, cloud APIs, and plugins网关、云端 API、插件
Governance output治理输出
Enforcement & Evidence处置与证据
Destination触达目标
External Systems外部系统
Core insight核心洞见
Govern context not just content. 关注语境 而非内容
CI dynamic privacy judges by actor × audience × purpose—not keywords alone. Secrets stay out of model traffic; task structure is preserved.CI 动态隐私按行动者 × 受众 × 目的判定，而非仅靠关键词——机密不出模型，任务结构保留。
Actor行动者
Audience受众
Purpose目的
Content内容
Norms规范
Trade secret scenario商业机密场景
Draft a follow-up email to John Smith, CEO of Acme Corp, on the Project Phoenix acquisition. Cover our 12.8× EV/EBITDA multiple, the three-tier pricing model (Tier-1 margin 42%, Tier-2 35%), and licensing for undisclosed catalyst CX-7 (formula C₁₂H₁₈O₄). Account AC-4829.请基于 Project Phoenix 并购案谈判纪要，起草一封发给 Acme 公司 CEO John Smith 的跟进邮件，说明我们提出的 12.8 倍 EV/EBITDA 估值、三层阶梯定价模型（Tier-1 毛利率 42%、Tier-2 35%），以及尚未公开的催化剂技术 CX-7（分子式 C₁₂H₁₈O₄）的授权条款，参考账户 AC-4829。
Draft a follow-up email to [PERSON_1], CEO of [ORG_1], on the [PROJECT_1] acquisition. Cover our [MULTIPLE_1] EV/EBITDA multiple, the [PRICING_MODEL_1], and licensing for undisclosed catalyst [TECH_1] (formula [FORMULA_1]). Account [ID_1].请基于 [PROJECT_1] 并购案谈判纪要，起草一封发给 [ORG_1] CEO [PERSON_1] 的跟进邮件，说明我们提出的 [MULTIPLE_1] 估值倍数、[PRICING_MODEL_1]，以及尚未公开的催化剂技术 [TECH_1]（分子式 [FORMULA_1]）的授权条款，参考账户 [ID_1]。
The LLM drafts the same quality email—your employee sees the final output with real values restored. Productivity unchanged. Zero secret leakage to the model provider.大模型会生成同样出色的邮件——员工看到最终输出时，所有真实数值均已恢复。生产力丝毫不受影响，向模型服务商零机密泄露。
One governance console—visibility, verdicts, and evidence before every request leaves.统一治理视图：每一次 AI 请求出站前，可见、可判、可举证。
Powered by Policy Engine small models—privacy, security, and compliance enforced before transmission, with versioned rules and explainable verdicts.由 Policy Engine 小模型驱动，在传输前落实隐私、安全与合规；策略版本可追溯，裁决可解释。
Employees, copilots, APIs, and agents in one view员工、Copilot、API 与智能体统一呈现
One console for employee, copilot, API, and agent traffic. Risks, reviews, and outcomes in a single view.一个控制台集中查看员工、Copilot、API 与智能体流量，风险、复核与处置结果同源可查。
Governance value治理价值Shadow AI and agent risk surfaced early—before it spreads.影子 AI 与智能体风险及早发现，避免扩散。
Finish reviews within SLAs you set按自设 SLA 完成审查
Configurable timeouts for compliance, privacy, and safety. Human review fallback when needed.合规、隐私与安全审查可配置时效，必要时支持人工复核兜底。
Governance value治理价值Adopt AI safely—without compliance becoming a bottleneck.在保证安全的前提下加快 AI 落地，避免合规拖累业务。
Risk-appropriate actions按风险处置
Map P0–P3 to risk-appropriate actions for your environment—least-privilege for agents and tools.将 P0–P3 映射为放行、脱敏、拦截等匹配动作；智能体与工具遵循最小权限。
Governance value治理价值High-priority risks get the right response—without blocking everyday traffic.高危优先处置，日常流量少误拦。
Tune policy, knowledge, thresholds, and RBAC配置策略、知识库、阈值与角色权限
Train regulations, corporate bylaws, and industry standards into small models. Versioned, role-controlled, threshold-tunable.将法规、企业条例与行业规范训入小模型，支持版本管理、角色权限与阈值配置。
Governance value治理价值Internal rules become live controls across every AI call path.内部制度转化为全链路 AI 调用的实时管控。
Plug into DLP, SIEM, and IAM you already run对接现有 DLP、日志与身份系统
Integrate DLP, moderation, threat intel, and knowledge bases by policy. Adds the AI governance layer on top of SIEM, DLP, and IAM.按策略对接 DLP、内容审核、威胁情报与知识库，在现有日志、DLP 与身份系统上叠加 AI 治理层。
Governance value治理价值Extend your existing security stack—no rip-and-replace.在既有安全体系上补齐 AI 治理能力，无需推倒重来。
Measured performance可量化的识别与执行能力
95%+ detection accuracy, ≤5% false positives, low cost, 10K+ concurrency—benchmark figures from internal testing; validate in your POC environment.95%+ 检测准确率、≤5% 误报率、低成本、万级并发（内部基准测试数据，请在试点环境验证）。
Governance value治理价值High accuracy and low false positives—governance at scale without slowing the business.准确识别、低误报，策略可规模化落地而不拖慢业务。
Sovereign control for strict environments. Hybrid speed for distributed teams.严监管本地主权，分布式团队混合敏捷。
Subnet gateway takeover for in-domain control, or multi-protocol Guard for unified local review.子网网关接管实现域内控制，或 多协议 Guard 统一本地审查。
Sovereign AI Firewall
Enterprise AI traffic企业 AI 流量
Subnet AI Gateway子网 AI 网关Certificate / gateway takeover证书 / 网关接管
Small models · In-domain enforcement小模型 · 域内处置
AI Traffic Guard
Protocol AProtocol B+
Multi-protocol Guard多协议 GuardUnified local traffic review本地统一流量审查
Same governance · Lighter footprint同一治理语义 · 更轻部署
Sovereign AI FirewallEnterprise subnet gateways take over AI traffic inside the
network.企业子网网关在本网接管 AI 流量。
AI Traffic GuardParses multiple AI protocols locally for unified traffic
review and management.本地解析多种 AI 协议，统一完成流量审查与管理。
Sovereign AI Firewall
For regulated enterprises that need data to stay in-domain面向需要数据不出域的严监管企业
Private AI security gateway. Enterprise subnet gateways take over in-domain AI traffic.私有化 AI 安全网关，由企业子网网关接管域内 AI 流量。
Data in-domain数据不出域Local inference本地推理Certificate takeover证书接管Audit-ready审计就绪
Data stays in the enterprise boundary.数据不出域。
Identify, judge, and enforce inside the network.在本网识别、判断与处置。
Semantic-consistent desensitization.语义一致性脱敏。
Rewrite sensitive data while preserving task
intent.保留任务意图，替换敏感信息。
Subnet gateway takeover.子网网关接管。
Intercept AI traffic at subnet boundaries before it
leaves.子网边界拦截离网前 AI 流量。
Local small-model inference.本地小模型推理。
Models near traffic for low latency, predictable cost,
and audit evidence.模型贴近流量，低延迟、可预测成本与证据沉淀。
AI Traffic Guard
For distributed teams using cloud AI across many exits面向多出口、分布式使用云端 AI 的团队
Lightweight Guard parses multiple AI protocols locally for unified review and management.轻量 Guard 本地解析多种 AI 协议，实现统一审查与管理。
Lightweight Guard轻量守护Cloud upgrades云端升级Remote workforce远程办公Cloud AI云端 AI
Multi-protocol parsing.多协议解析。
One local layer for diverse AI client and SaaS
protocols.本地一层解析多种 AI 客户端与云端协议。
Cloud intelligence keeps pace.云端能力持续跟上新风险。
Cloud models and rule packs iterate as tools and
regulations evolve.随新工具与监管要求迭代云端模型与规则包。
API calls and Plugin integration.支持 API 调用与 Plugin 接入。
Govern AI across distributed teams, remote sites, and
cloud apps with the same policy engine.以同一策略引擎，覆盖分布式团队、远程站点与云端应用的 AI
治理。
Same governance semantics as the firewall.与防火墙共享治理语义。
Same risk levels, actions, approvals, and evidence model
as Firewall.与防火墙共享风险等级、处置、审批与证据模型。
What changes when AI traffic is governed.AI 流量被治理后会发生什么变化。
Before治理前
Deal terms and client emails reach public copilots unnoticed.交易条款与客户邮件悄然进入公共协同助手。
With Mode IOMode IO 治理后
Detect trade secrets in context; handled according to risk level, with every action logged.识别商业机密，按风险等级处置，全程留痕。
Before治理前
Over-scoped agents and hijacked prompts exfiltrate data with no runtime block.智能体越权、提示词劫持可外泄数据，运行时无法拦截。
With Mode IOMode IO 治理后
Least-privilege enforcement, ASI blocking, P0–P3 audit trail.最小权限执行、ASI 拦截、P0–P3 审计留痕。
Before治理前
Regulated data exits via AI prompts with no proof under regulations like PIPL and GDPR.受监管数据经 AI 提示词出境，缺乏 PIPL、GDPR 等多项法规证据。
With Mode IOMode IO 治理后
Jurisdiction and purpose checks. Risk-appropriate enforcement with audit evidence.辖区与目的审查，按风险分级处置并留证。
Platform module平台模块
Cross-Border Compliance跨境数据合规
GDPR, CCPA, PIPL, and other major regulations. Deploy with Firewall, Guard, or standalone.覆盖 GDPR、CCPA、PIPL 等主要法规，可与防火墙或 Guard 一并部署，亦可独立扩展。
Learn more了解详情 → Underlying technology底层技术Policy Engine策略引擎
Trains regulations, corporate bylaws, and industry standards into small models. Powers every module—or deploy standalone for employee policy compliance.将法规、企业条例与行业规范训入小模型，驱动各模块；亦可独立部署，监管员工制度遵从。
Learn more了解详情 →From small-model judgment to enforceable evidence.从小模型审查到可执行证据链。
Each decision produces rationale, action records, and audit-ready evidence.每次判断形成审查依据、处置记录与可审计证据。
Explainable review可解释审查
Policy match, risk category, sensitive span, and recommended action.命中策略、风险类型、敏感片段与推荐处置。
Action record处置记录
Enforcement outcomes stored with event IDs and user context.处置结果关联事件 ID 与用户上下文留存。
Audit retrieval审计检索
Reconstruct who requested what, which policy fired, and why.还原请求人、命中策略与系统处置原因。
Configurable SLA可配置时效
Define review timeouts per compliance, privacy, or rewrite workflow.按合规、隐私或改写流程配置审查时效。
P0–P3 priorityP0–P3 风险优先级
Map P0–P3 to risk-appropriate actions. Configurable by context.将 P0–P3 映射为按风险等级匹配的动作，可按场景配置。
Human fallback人工复核
Routes to human review on timeout or low confidence.超时或置信度不足时转入真人复核。
Knowledge base versioning知识库版本管理
Upload regulations and internal rules with version markers for traceability.上传法规与企业制度，版本标记保留追溯能力。
Semantic firewall语义防火墙
ASI-01 intent allowlists and block or semantic substitution for sensitive data.ASI-01 意图白名单，敏感数据拦截或语义替换。
RBAC and approvals角色权限与审批
Admin, auditor, developer, and custom roles manage thresholds and approval routes.管理员、审计员、开发员与自定义角色管理阈值与审批路径。