AI Registry

Last Updated: September 10, 2026 | By Mihail Sebastian | AI Dictionary

An organization's inventory of approved AI assets: models, applications, agents, and MCP servers, with ownership, approval status, and review dates.

What is an AI Registry?

An AI registry is an organization’s inventory of approved AI assets: the models, applications, and agents in use, each recorded with an owner, an approval status, and a risk classification.

It is the counterpart to shadow AI: the registry defines what “approved” means, and anything outside it is by definition unaccounted for. Increasingly the inventory covers not just models but the tools agents reach, including MCP servers.

How an AI Registry Works

A registry runs on a lifecycle. At intake, whoever wants a new model, application, or tool submits it with the intended use case. At review, a governance function assesses the risk, checks the vendor, and classifies the asset.

Approval attaches conditions: approved for internal data only, approved with human review of outputs, or rejected. Each entry names an owner who answers for the asset, and periodic re-review catches entries whose risk changed since approval.

The registry earns its keep only if the path through it is fast. When approval takes months, teams route around it, and the inventory stops describing reality.

Regulation pushes in the same direction. The EU AI Act obliges organizations to know which AI systems they operate and to classify them by risk, and a maintained registry is where that evidence naturally lives. It also hands an AI audit its starting list.

AI Registry vs AI Gateway

The registry is the record of what is approved; the AI gateway is the runtime checkpoint that enforces it. One is a policy artifact consulted by people, the other a piece of infrastructure sitting in the traffic path.

They work as a pair: the registry decides, the gateway executes the decision on every request.

CriterionAI RegistryAI Gateway
RoleRecord of what is approved and who owns itEnforcement of policy on live traffic
When it actsAt intake and review, before useAt request time, during use
What it holdsOwners, approval status, risk classLogs, quotas, credentials, guardrail rules
Without itNobody knows what AI is in useRules exist on paper, nothing applies them

Example of an AI Registry

A data science team wants an external LLM API for contract summarization. They file an intake request naming the provider, the data involved, and the use case.

The review board classifies the use as limited-risk but flags that contracts contain customer data, so approval comes with a condition: access only through the company gateway, with logging on. The entry lists the team lead as owner and sets re-review in twelve months.

Six months later a colleague searches the registry, finds the approved API, and reuses the entry instead of putting a new vendor through review from scratch.

Related AI terms: AI Governance · Shadow AI · MCP · AI Audit · Compliance

Did you like the AI Registry gist?

Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.

Written by

Mihail Sebastian

Mihail Sebastian

Editor, AI Guv

Mihail works in AI and writes about artificial intelligence topics for people who need to understand it without building it. He comes from more than 20 years of product design in startups.

Read the Governor's Letter

Stay ahead with Governor's Letter, the newsletter delivering expert insights, AI updates, and curated knowledge directly to your inbox.

By subscribing to the Governor's Letter, you consent to receive emails from AI Guv.
We respect your privacy - read our Privacy Policy to learn how we protect your information.

Browse All AI Terms A–Z

Every term in the dictionary, in alphabetical order. Jump to a letter or scroll the full list.

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

S

T

U

V

W

X

Y

Z