How to Discover & Store AI Assets in ServiceNow

Businesses didn't wake up one day and find themselves surrounded by AI. It crept in. A model here, a pilot there, a proof-of-concept running under someone's desk, or a chatbot built by a curious manager. For many organisations, they are using AI everywhere, but they can't actually see it.

We were having this exact same conversation in our internal AI Practice the other day. We know people have started using AI in many creative and impactful ways, but we can't really see it, let alone govern it. And when I say govern it, it goes both ways. For those low-risk, impactful use cases we want to circulate that as far and wide in the business as we can, and then for those high-risk use cases, if there are any, we want to make sure we have visibility and control.

As part of the Zurich release, ServiceNow have released the AI Control Tower which aims to address that issue. It introduces new capability that expands the traditional Configuration Management Database (CMDB) to create a robust, CSDM-aligned AI asset inventory that treats AI as a first-class asset, with structure, lineage, and traceability.

But how does it work? In this article I'm going to explore the architecture and discovery mechanisms that ServiceNow are using to tackle this governance issue.

An Overview of the AI Inventory and Control Tower

AI has outpaced traditional configuration management. The CMDB didn’t know what an AI agent was, couldn’t model relationships between a model and the systems using it, and had no idea how to track prompts or usage patterns. AI Control Tower is ServiceNow’s answer to all of this. It provides a home for AI assets. A space where organisations can finally unify:

• Where their models live

• How their agents behave

• Which datasets feed them

• How they are used

• How they should be governed

This inventory expands the CMDB through the introduction of new classes specifically designed to track key AI components:

1. AI Systems: These are the deployed applications or agents.

2. AI Models: These represent the underlying intelligence, such as foundation models like GPT-5 or fine-tuned custom models.

3. AI Digital Assets: These include related components, primarily data artefacts like training datasets, knowledge bases, or AI “tools”.

Alignment with CSDM v5

ServiceNow have shaped the AI Inventory so it fits neatly into the CSDM v5 model, meaning your AI assets sit exactly where the rest of your technology estate already lives. It keeps everything consistent and gives you that clean line of traceability from a business capability right down to the AI powering it.

Here’s how the main pieces slot in.

AI Systems

Think of these as the actual AI solutions running in your world. Some are things you’ve built yourself on ServiceNow. Others live entirely externally, like Azure AI Foundry or AWS Bedrock. ServiceNow splits them into two types so you know what’s internal and what’s hosted elsewhere:

• AI Application for anything you’ve developed directly on the platform.

• AI Function for externally hosted AI, which is where most hyperscaler services end up.

Anything discovered from external sources lands as an AI Function that you can link straight back to the business applications it supports. It keeps the relationships clear without you having to stitch it all together by hand.

AI Models

These are the brains behind the systems. Zurich introduces a proper AI Model record in the CMDB so you can actually catalogue models like GPT-5, Amazon Titan, or anything custom your teams have fine-tuned. Discovery creates one record per model and relates every AI System using it, so you finally get a clean view of where each model is deployed.

AI Digital Assets

This is where all the supporting material lives: datasets, prompt libraries, knowledge bases, vector stores, tools, and so on. These aren’t deployable systems, so ServiceNow treats them as assets rather than CIs. They still stay fully linked to the AI Systems and AI Models that use them, but they don’t clutter up your configuration hierarchy. It keeps things tidy while still giving you a proper inventory of everything that fuels your AI solutions.

Azure and AWS Connectors

The whole ability to auto-populate the AI Inventory really comes down to the native discovery content ServiceNow introduced in Zurich. This isn’t generic cloud scanning. It’s proper, hyperscaler-aware discovery designed specifically for modern AI services.

Discovering Azure AI Foundry

ServiceNow has built OOTB patterns to understand Azure AI Foundry. When the platform connects into your Azure subscription, it can spot Azure OpenAI instances straight away. Each one gets onboarded as an AI System CI and classified as an AI Function because it’s an externally hosted service.

From there, the platform reaches into Azure’s management APIs to pull back the real detail. It lists the deployments inside the resource, so models like GPT-4 or GPT-5 get recorded automatically as AI Models in the CMDB. If you’ve built more advanced agentic capability in Azure, such as a copilot or an orchestration layer, that gets recognised too. The agent becomes an AI System and anything it relies on, like a Cognitive Search index, lands as an AI Digital Asset.

You even get usage insights. Prompt counts, inference activity, that kind of thing. It all feeds into a richer picture of how the service is being used.

Discovering AWS Bedrock

Zurich also includes a full discovery flow for Amazon Bedrock. ServiceNow connects through AWS APIs, finds any Bedrock Agents you’ve deployed, and drops them into the CMDB as AI Functions.

It also works out which foundation model each agent runs on, whether that’s Claude, Titan, or anything else. Those models get logged as AI Model records, and the platform handles reconciliation so you don’t end up with duplicates if multiple agents use the same one.

Any tools the agent is configured to call, or any knowledge bases or vector stores attached to it, are captured as AI Digital Assets. And again, usage metrics come through from Bedrock’s activity logs, giving your AI governance teams something concrete to monitor inside the Control Tower dashboards.

In short, the Zurich release finally gives you a simple way to discover AI capability across Azure and AWS without needing custom setups. It just works, and it gives you the visibility most organisations have been missing.

Plugins, Licensing, and Setup

Leveraging all this new discovery capability does come with a few setup steps, but the good news is that ServiceNow have kept things fairly straightforward.

What you need to have in place

To switch on AI asset discovery, your organisation needs a handful of things enabled:

• AI Control Tower licensing: AI Control Tower is its own licensed product, usually owned by an AI Centre of Excellence or Governance team. If your organisation has subscribed to it, you’re already most of the way there.

• AI Governance Core Plugin: This is the foundational piece. It brings in the extended CMDB model for AI Systems, AI Models, and AI Digital Assets, all structured properly under CSDM v5.

• AI Discovery Integration Plugin (sn_ai_disc): This is the Store app that contains the out-of-the-box discovery patterns for Azure AI Foundry and AWS Bedrock. If you have the AI Control Tower SKU, this plugin arrives automatically.

What you don’t need

A lot of people assume you need a full stack of other ServiceNow products to make this work. You don’t.

• You don’t need ITOM licences.

• You don’t need a separate IntegrationHub licence.

• You don’t need to build any custom discovery patterns yourself.

• And you definitely don’t need a MID Server, because everything runs over simple outbound HTTPS calls directly to Azure or AWS.

However, it is worth keeping in mind that AI Control Tower doesn’t sit in isolation. Its AI-specific capabilities work neatly alongside the wider platform features you may already use at an organisational level, such as strategic planning or risk management.

Getting everything connected

Once the plugins are in place, the only real job for admins is to set up the cloud credentials. That might be a Service Principal in Azure or an IAM role or access keys in AWS Bedrock. As long as the credentials have read access to the AI services, you’re good.

From there, Guided Setup takes over. The AI Control Tower’s Discovery Setup page walks you through plugging in the credentials and testing the connection. After that, the platform starts querying your cloud environments automatically and begins filling out the AI System, AI Model, and AI Digital Asset tables for you.

Conclusion

The AI Control Tower in the Zurich release finally gives organisations what they’ve been missing: a clear view of all the AI wandering around their estate. Instead of guessing where all of your AI capability actually lives, the platform now brings it together into one place. Whether you’ve built something directly on ServiceNow or teams have spun up solutions in Azure AI Foundry or AWS Bedrock, it all gets discovered automatically and pulled into a single, consolidated inventory inside the CMDB.

Because it uses the CSDM v5 structure, each piece lands in the right place. AI Systems show up as CI Functions, models are catalogued as proper CMDB Models, and datasets or tools sit neatly as linked Assets. The end result is a governed inventory that actually makes sense. It becomes possible to connect risk, compliance, and operational checks directly to the AI that’s powering your business, ensuring the technology doesn’t run ahead of the guardrails you should already be relying on.