AutoML on Databricks as of May 2026

[nepiskopos]

On Azure Databricks, I have been creating AutoML experiments on May 15th 2026, using a Single User compute with Databricks runtime 17.3 LTS for ML, which I have assigned to myself.

I try to register one of the trained ML models in a schema, under a unity catalog, using the Databricks web UI.

However, no model is registered. A placeholder with the assigned model name gets created in the schema under Models, but no model version is registered.

This procedure was working two weeks ago.

I have ran the following SQL command
SHOW GRANTS `your_email@domain.com` ON SCHEMA <your_catalog>.<your_schema>;
and I verified I have CREATE MODEL permissions on the specific schema.

I have also ran the following SQL command
SHOW GRANTS `your_email@domain.com` ON CATALOG <your_catalog>;
and I verified I have both USE CATALOG and USE SCHEMA permissions on the specific catalog.

If I try to register the model using MLflow, it succeeds.

Why is this suddenly happening?

[Ashwin_DSA]

Hi @nepiskopos,

From what you’ve described, this does not look like a Unity Catalog permissions issue. The main reason is that the target registered model object is being created in the schema, but the model version is not. That usually points to a failure in the UI registration flow after the model placeholder has been created, rather than a problem with CREATE MODEL / USE CATALOG / USE SCHEMA itself.

The other strong signal is that registration works when you do the same thing through MLflow. If the same model can be registered successfully with MLflow into the same catalog and schema, that makes a permissions problem much less likely and suggests the issue is specific to the Databricks web UI path.

So this is most likely a recent UI-side regression or failure in the backend path used by the UI for model version creation. In other words... the UI can create the model entry , but it is failing on the follow-up step that should create the first model version

That would also explain why this worked a couple of weeks ago and is now suddenly failing, without any obvious change in your grants.

For now, the practical workaround is to continue registering the model through MLflow, since that path is succeeding in your workspace.

You may also want to raise a support ticket with the relevant details for the product teams to investigate.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.