ZTNA for AI

Extend Zero Trust to Every AI Agent

ZTNA for AI gives each agent a governable identity and enforces least privilege access to models and tools through a single, inline AI Gateway. Every request is attributed and recorded, giving organizations control and audit evidence without forcing developers to rewrite existing agents. 

Schedule a Demo

Benefits of ZTNA for AI

AI agents can act across enterprise systems using the access of the people or services that launch them. AppGate ZTNA for AI gives security leaders, architects and development teams an identity-based framework for governing that access while preserving existing agent workflows.
  • Faster, provable control: Organizations can establish enforceable policies for a previously ungoverned identity class without slowing agent adoption.
  • Reduced blast radius: Least-privilege, deny-by-default entitlements limit which models and tools each agent can access and constrain what a hijacked or misbehaving agent can do.
  • Audit-ready evidence: Every agent request produces a fully attributed record that security and compliance teams can use without reconstructing activity after the fact.
  • No developer disruption: Integration changes a base URL rather than application code, so existing agents and developer workflows can continue unchanged.
  • Resilient by design: The Gateway validates signed tokens and enforces their entitlements locally, so a control-plane outage affects new-token issuance and refresh rather than live enforcement. 
Specialist maintenancing ai systems

Govern Every Agent Request Through One Inline Gateway

Content guardrails focus on what a model says. Agentic systems also need access controls that govern what an agent is allowed to do. ZTNA for AI registers each agent as a revocable identity and routes every agent-to-model and agent-to-tool request through a single inline AI Gateway. 

Establish First-Class Agent Identity

Each agent becomes an identifiable, revocable principal, authenticated through the enterprise identity provider.

Apply Least Privilege Per Model & Tool

Entitlements carried in a signed token determine access for each agent identity on every request.

Keep Enforcement Inline

The AI Gateway validates the token, enforces its entitlements as traffic passes through and produces a fully attributed audit record.

Integrate Without Rewriting Applications

Changing the base URL brings existing agents under governance without re-engineering the workflows they support.

Deploy in Your Own Cloud

The customer-controlled deployment does not depend on a vendor-hosted control service. 

How ZTNA for AI Works

ZTNA for AI gives every AI agent a governable identity and routes each request through a single inline AI Gateway. The Controller authenticates the agent through your identity provider (IdP) and issues a signed, short-lived token. The Gateway enforces that token's entitlements locally and records every request.

AI agents

Copilots, apps and automations

Access request

Signed token

Controller

Authenticates via IdP, mints a signed, short-lived token

Enterprise IdP

Request + token

2

AI Gateway

Enforces entitlements on every request and calls AIDR where enabled

Brokered

  • Models

    LLM endpoints

  • MCP tools

    Enterprise systems

Blocked inline

Requests outside entitlements

Inspect

Redacted

3

AI Detection & Response (AIDR)

Runs on AppGate ZTP. Redacts sensitive data and alerts

Verdicts, alerts

1

Audit record

Every request, response and verdict, fully attributed

Resilient by design: the Gateway enforces entitlements locally, with no callback to the control plane, and AIDR runs on AppGate ZTP rather than the control plane. A control-plane outage pauses new token issuance and refresh, while enforcement for active agents continues.

Request flow: agents authenticate through the Controller and your IdP and receive a signed, short-lived token. The AI Gateway validates the token and checks entitlements on every request, blocking anything not entitled. Where enabled, AIDR inspects the request, redacts PII and other sensitive data and raises an alert before the Gateway brokers it to models and MCP tools. Requests, responses and verdicts are written to a fully attributed audit record.
  • 1 Observe

    Log every agent-to-LLM and agent-to-MCP request and response with its detection verdict and attribute it to an identity, model, tool, action and token usage.

  • 2 Control

    Enforce least-privilege, deny-by-default entitlements for each model and tool request using identities authenticated by the enterprise identity provider.

  • 3 Protect

    Turn on AI Detection & Response (AIDR) per agent in its entitlements to redact PII and other sensitive data inline and alert your team before it reaches a model or tool.

Featured Resources

Coworkers doing brainstorming solving tasks in ai
Datasheet
AppGate ZTNA for AI

Learn how ZTNA for AI treats every AI agent as a governable identity, enforcing least-privilege access to each model and tool while capturing a fully attributed audit trail, all without rewriting your agents’ code.

Learn More
Whitepaper
Access Control Is the Real Test of AI Sovereignty
Webinar
AI Is Already on Your Network
Article
What the OpenAI Incident Reveals About the Future of Cyber Defense

Bring AI Agents Under Identity-Based Control

See how AppGate ZTNA for AI can help your organization establish agent identities, enforce least-privilege access and produce audit-ready evidence without rewriting application code.

Book a Demo