What's new in AI Gateway v3.4 - September 2026 release

What is AI Gateway?

AI Gateway is ServiceNow's runtime enforcement layer for MCP (Model Context Protocol) connections in AI Control Tower. It sits between your AI agents and the external tools they call, governing what reaches agents, securing every connection, and giving you complete visibility into everything happening across your MCP server landscape.

AI Gateway is part of AI Control Tower and is configured through the AI Control Tower interface. Whether you're an AI Steward building a governed MCP program from scratch, a CISO who needs runtime control and audit evidence, or a developer connecting an agent to external tools, this article covers everything AI Gateway does today.

💡 A Foundation Built to Grow

This release consolidates governance, security, and observability into a single governed enforcement layer inside AI Control Tower. If you're already using AI Gateway, you'll notice the refreshed interface right away - but the bigger change is under the hood, in a unified architecture that separates configuration in AI Control Tower from enforcement at the gateway.

That separation matters. It's what lets AI Gateway apply one consistent set of policies across every MCP server and tool, no matter how that server entered your inventory — and it gives the platform the flexibility to take on new capabilities without reworking how governance is configured.

ServiceNow may make changes to planned features based on market feedback and customer priorities.

⚠ Upgrading from a previous release?

One change from this release is visible right away: the AI Gateway proxy URL format has changed.

New format: https://<instance-url>/sncapps/aigw/mcp/<mcp-server>

Previous format: https://<instance-url>/sncapps/awh/<mcp-server>/mcp

Existing MCP and CIMD registered clients can be edited to update their configuration from the AI Gateway Setup tab. Review any client integrations pointing to the previous URL format and update them before agents lose their connection.

View AI Control Tower release notes

📑 Looking for this release

This is the AI Gateway release documented here. In the ServiceNow Store, it appears as version AWH for AI Control Tower v3.4.

Govern Control Observe Who It's For

📋 Govern

The Govern pillar ensures every MCP server in your enterprise has been reviewed, approved, and tracked through its full lifecycle before any agent can use it.

1. MCP Server and Tool Catalog

AI Gateway maintains a single, governed inventory of every MCP server in your enterprise, regardless of how it was registered. Servers from AI Agent Studio, MCP Server Console, and the built-in MCP catalog all surface in one place in AI Control Tower, and every server comes with a full list of the tools it exposes.

Agents can only connect to servers that are approved and active. Unapproved servers are blocked in AI Agent Studio with no developer workaround possible.

Four ways an MCP server enters your inventory

Auto-discovery from AI Agent Studio: When a Product Owner configures an MCP server in AI Agent Studio, it automatically syncs to AI Control Tower for AI Steward review. No separate intake step required.

MCP Server Console: Teams managing external integrations can register servers directly through MCP Server Console. These surface in the same inventory alongside auto-discovered servers.

MCP catalog import: AI Stewards and Product Owners can browse and import servers from the built-in MCP catalog. Selecting a server auto-populates its URL, authentication type, and exposed tools — no manual entry required.

Manual entry: AI Stewards and Product Owners can import MCP servers not auto-discovered or available in the MCP catalog import manually.

Inventory view in AI Control Tower showing the MCP servers list with name, asset type, state, status, managed status, and managed-by columns

Who this affects: AI Steward , Product Owner , Developer

2. Lifecycle Management and Approvals

Every MCP server moves through a defined lifecycle in AI Control Tower: from intake through review, approval, active deployment, and eventual deprecation. The AI Steward owns this process end to end through the Asset Approval Playbook, the same workflow used for other AI assets on the platform.

When an AI Steward approves a server, it becomes available to agents automatically. Approval status flows directly into AI Agent Studio so developers can see which servers are approved and which are still under review, with no manual coordination required. Policies set at the tool level inherit automatically across every agent that calls that tool.

MCP server lifecycle states

New: entered inventory, not yet reviewed · In Review: approval workflow open · Approved: cleared for deployment, AI Gateway provisioned · Active / Deployed: live and available to agents · Deprecated: retired, full history preserved

MCP server detail record, Lifecycle tab showing approval status and history log

Who this affects: AI Steward , Product Owner , Compliance & Security

⏯ Control

The Control pillar ensures every connection is authenticated, every tool is scanned before use, and sensitive data never reaches an agent it shouldn't.

1. Authentication and Connectivity

Every MCP connection is authenticated through AI Gateway. There are no direct agent-to-server connections, no unscoped tokens, and no credentials stored on the agent side. AI Gateway verifies agent identity, issues scoped short-lived OAuth 2.1 tokens on every connection, and rotates server credentials centrally. Server credentials are never exposed to the agent.

Any MCP client, whether in AI Agent Studio, Copilot Studio, or a developer tool like Visual Studio Code, connects and becomes compliant through a guided setup in AI Control Tower.

Supported client types

MCP client: for agent platforms connecting via the standard MCP protocol

CIMD client: for developer tools using Client Identity Metadata Discovery for automated, frictionless registration without manual credential entry

AI Gateway tab on an MCP server record, Setup sub-tab showing Gateway URL, authorization endpoint, token endpoint, and MCP client integration list

Who this affects: AI Steward , Developer , Compliance & Security

2. Runtime Policy Enforcement: Pause

AI Gateway enforces access policies in real time across every agent-tool interaction. If a server behaves unexpectedly, is flagged by a security review, or needs to come offline for any reason, you can pause it — globally across all connections, or individually for a specific server — without any code changes.

Other servers stay running. Configurations are fully preserved while a server is paused. When the situation is resolved, you restore with one click — no setup needs to be repeated.

WITHOUT PAUSE CONTROLS

Stopping a specific server means taking down every agent that depends on it. A targeted security response becomes a broad service disruption.

WITH AI GATEWAY

Pause any server globally or individually from AI Control Tower. Other servers keep running. Restore in one click when the situation resolves.

Summary: without pause controls, stopping one server risks disrupting every dependent agent. With AI Gateway, you pause the affected server only, and restore it in one click once resolved.

Settings > AI Gateway overview showing "Connections are currently running" status with a global Pause button and a per-server connection list with individual Pause buttons

Who this affects: AI Steward , Compliance & Security

3. Tool Scanning and the Tool Catalog

Before a tool can be activated, AI Gateway scans it to determine whether it poses a potential security threat. Threat categories are based on the OWASP MCP Top 10, the ten most critical MCP-related security vulnerabilities as defined by OWASP.

Two instance-level settings automate how AI Gateway handles the results. Both are enabled by default and recommended for production environments: automatically deactivating any tool found vulnerable during a scan, and automatically activating new tools fetched after a server has been approved.

Scan status values

Completed · Failed · Pending · Timeout

Tool Scan.png

Tools sub-tab on the AI Gateway tab showing tool name, description, scan status, and threat category columns, with the Settings > AI Gateway > Tools tab showing the two automation toggles

Who this affects: AI Steward , Compliance & Security

4. Data Sensitivity Check

The Data Sensitivity Check inspects tool inputs and outputs for sensitive data. When enabled, it blocks the response before it passes between an AI agent and the MCP server — if the response contains a data type your policy covers, the entire response is stopped.

AI Stewards activate the Data Sensitivity Check policy from the Policies sub-tab and configure which data types are in scope, at the server level or the individual tool level.

Policies sub-tab on an MCP server record showing the Data Sensitivity Check policy panel with data type selection

Who this affects: AI Steward , Compliance & Security

📊 Observe

The Observe pillar gives AI Stewards full operational visibility into every AI Gateway connection, with no agent-side instrumentation required.

1. Operational Visibility

Every AI Gateway call is captured, traceable, and visible in AI Control Tower from the day a server is approved and active. All telemetry is captured at the gateway boundary — no instrumentation needed on the agent side.

AI Stewards spot underperforming servers, unauthorized access attempts, and connection health issues before they become incidents.

What's recorded per server, tool, and client

Success rate: percentage of tool calls that completed successfully vs. failed

Latency: P50, P90, and P95 response times per tool, so you can identify slow tools before users feel the impact

Connection attempts: authorized access attempt details per MCP server and client, including client identity

Metrics are available from the Monitor tab on any MCP server record. The MCP servers overview in AI Control Tower also shows a time-series chart of successful vs. failed access attempts across all servers.

Monitor tab on an MCP server record showing the tools observability metrics table (tool name, requests, success rate, latency P50/P90/P95) with the access attempts time-series chart above it

Who this affects: AI Steward , Product Owner , Compliance & Security

👥 Who It's For

Role	What they use AI Gateway for	Key capabilities
AI Steward	Governing the MCP server program end to end, from intake through retirement	Approval workflows, tool catalog management, pause controls, observability metrics
CISO	Ensuring no AI agent reaches a tool that hasn't been vetted, with an instant response capability when something goes wrong	Pause controls, Data Sensitivity Check, tool scanning, access attempt logs
Enterprise Architect	Designing a governed MCP architecture that works across agent platforms, not just ServiceNow	Platform-agnostic Gateway URL, OAuth 2.1 authentication, centralized credential management, CIMD client support
Developer / Builder	Connecting agents to MCP tools through a governed path, knowing which tools are approved and ready to use	Gateway URL (replaces direct server connection), tool catalog browsing, CIMD frictionless registration, lifecycle status visible in AI Agent Studio

📚 Related Resources

Implementation Guide

AI Gateway Implementation Guide (Comming soon!)

Step-by-step configuration for every setup workflow: intake, approvals, client registration, security controls, and observability.

FAQ

AI Gateway FAQ

Answers to common questions about AI Gateway setup, governance, and MCP server management.

|
|

Product Documentation

AI Gateway Product Documentation

Official docs covering AI Gateway architecture, configuration options, and reference material.

AI Control Tower

AI Control Tower Product Documentation

Office docs covering AI Control Tower.