
Key Takeaways
- Zenity now secures AI at the network layer. Discovery of the AI in use across the organization, visibility into personal and unsanctioned accounts, and inline prevention of every AI activity.
- It connects to the SASE or secure web gateway organizations already run through standard proxy chaining, and covers AI in browsers, desktop apps, CLIs, and any other tool whose traffic passes through the gateway. Nothing to install on endpoints.
- Network activity lands in the same platform as every other agent Zenity secures, with the same Boundaries, findings, and investigation teams already know and love.
AI no longer lives in one place. Employees use chatbots in the browser, desktop assistants, and coding CLIs, from corporate and personal accounts alike. Some of that use is sanctioned and integrated. Much of it isn't, and security teams often don't know it exists.
What it all has in common is the network. Enterprises already route traffic through a SASE platform or secure web gateway that decrypts it and applies policy. The gateway sees every AI request and can match patterns in a single message, but AI risk rarely fits a pattern. Securing it requires understanding the interaction: what the user is trying to do, how the session has unfolded, whether a prompt is steering the model somewhere it shouldn't go, and whether a response should reach the user.
Today, Zenity extends its AI security platform to the network layer. The gateway an organization already runs hands its AI traffic to Zenity, which discovers the AI in use, attributes it to users, and evaluates every activity against the organization's Boundaries, blocking what policy forbids before it leaves or lands.
One Control Point for Every AI Interaction
Discover all AI use, including shadow AI. See which AI apps, services, and accounts employees actually use, sanctioned or not, and who uses them. Personal accounts, free tiers, and logged-out sessions never appear in platform APIs. At the network layer, they do, attributed to the corporate identity the gateway already authenticated. Shadow AI stops being a guess and becomes an inventory security teams can act on.
Prevent inline. Every activity is evaluated before it continues, in the context of the whole session rather than one message at a time. AI-based detections read intent, not just keywords, and stateful analysis follows the conversation. A user who pulls a customer list a few rows at a time across several prompts, or a prompt that gradually steers the model into revealing data, is caught even when no single message looks risky. A request carrying secrets, source code, or customer data is stopped before it reaches the AI provider, and a response that violates policy never reaches the user.
Enforce one policy. Boundaries apply exactly as they do to every other agent Zenity secures. The same Boundaries model that governs an enterprise agent now covers the same employee in a personal chatbot in the browser, a desktop assistant, or a CLI.
Live With a Few Gateway Settings
Proxy chaining is a standard feature of SASE platforms and secure web gateways. The gateway forwards selected traffic to another proxy for inspection. Turning on Zenity's network integration takes a handful of gateway settings: a chained proxy, the AI destinations to send to it, and a header that identifies the organization.
- Nothing on endpoints. No agent or browser extension to package, deploy, and maintain on every device.
- No new certificates. Devices already trust the gateway, so users see no changes.
- Broad coverage from day one. Every device whose traffic runs through the gateway is covered for each supported AI app.
- Fail-open by design. If Zenity can't be reached or can't reach a verdict, traffic flows as it did before. Only a deliberate policy decision blocks a request.
How It Works: The Gateway Routes, Zenity Decides
- The gateway decrypts AI traffic as it already does, and forwards the destinations you choose to Zenity over a secure, encrypted, authenticated connection, tagged with your organization and the user's identity.
- Zenity reads each activity in context: what was sent, who sent it, the app, the model, and the rest of the session.
- Zenity evaluates every activity against your policies and Boundaries in both directions: requests before they reach the AI app and responses before they reach the user.
- A deny stops the activity. An allow lets it through unchanged, and either way the activity is recorded in Zenity.
Network, API, or Endpoint: Where Each Fits
No single control point sees everything, which is why Zenity uses all three and brings them into one platform.
Network (new) | Platform integrations | Endpoint agent | |
|---|---|---|---|
Setup | A few gateway settings | Connect the enterprise tenant | Install and maintain software on every device |
Coverage | All AI traffic through the gateway: browsers, desktop apps, CLIs, personal accounts | Sanctioned enterprise tenants | Everything on the devices where it runs |
Inline prevention | On every request and response | Where the platform provides hooks | On AI activity on the devices where it runs |
Context | The traffic and its network context: what was sent, by whom, to which app, from which client and version (browser, desktop app, CLI) and from where (IP, region) | Deep: agents, configuration, knowledge, permissions, but no network context | Deepest: local tools, files, MCP servers |
Trade Offs | Sees only traffic the gateway routes and inspects; no view of local activity or of activity inside SaaS platforms | Blind to personal and unsanctioned use | Rollout and maintenance overhead that not every enterprise accepts |
The network is the fastest way to cover the most ground. It doesn't replace the depth of API and endpoint coverage, and it doesn't have to: all three feed the same findings, Boundaries and investigations.
In Practice: Sensitive Data in Unsanctioned AI
A support engineer pastes a customer's account details into a personal ChatGPT tab to draft a reply. A developer's AI coding CLI sends a config file with live credentials to a model provider. Neither tool is sanctioned, and neither goes through an integration the security team knows about.
Both requests pass through the organization's gateway, which forwards them to Zenity. Zenity matches them against the organization's Boundaries and denies them. The data never reaches the AI provider.
The employees see their requests blocked. The security team gets a finding for each, with the user, the app, and the Boundary that fired, and both tools appear in the AI inventory. Nothing was installed on either device.
Available Now, and Expanding
Network enforcement is available today with Zscaler Internet Access (ZIA). Support for more gateways and AI apps are on the way, and each one plugs into the policies customers already run.
Enterprises spent a decade putting a secure web gateway in front of every user. That investment is now an AI control point. The network already sees every AI interaction. Now Zenity can secure it.
Zenity customers who run a SASE platform or secure web gateway can ask their account team about network enforcement.
Not a customer yet? Book a demo, and we'll show you an AI request blocked before it leaves the network.
All ArticlesRelated blog posts

Intent-to-Kernel: Zenity Brings Agent-Aware Security to NVIDIA OpenShell
NVIDIA OpenShell is fast becoming the reference for running AI agents at scale. NVIDIA OpenShell™ Reference Architecture,...

Securing the Next Control Plane for AI Agents: Introducing Zenity for Sandboxes
Today we're introducing Zenity for Sandboxes: agent-aware security for every sandbox your agents run in. AI agents...

Zenity Now Secures Agents Across Your Microsoft Ecosystem
Microsoft has put AI agents into nearly every part of its ecosystem, and enterprises are using them, which makes...
Secure Your Agents
We’d love to chat with you about how your team can secure and govern AI Agents everywhere.
Get a Demo