Forum Discussion
Fabric + Purview (Data Classification)
Dears,
I am using Purview for data goverance (domains, data products, glossary terms, etc..) and data security (classification, sensitivity labels, DLP, etc..) of Microsoft Fabric
I found out that Purview , as opposite to other data sources, does not automatically classify the assets and its columns automatically while extracting the metadata via the data map funcionality
Can someone please let me know how are they doing the data classification via Purview for data assets in Fabric?
I would like to apply sensitivity labels on top of the assets or on top of the fabric items, but for that, I would need first to understand the classification according to the SITs and trainable AI classifiers that purview built-in
Thanks a lot,
Pedro
Hi @fabricpribeiro ,
Thanks for reaching out to the Microsoft Fabric Community forum.
At the moment, Purview’s Data Map scans for Microsoft Fabric do not yet perform content-based classification during the scan the way they do for some other sources like SQL or Databricks. Because of this limitation, the built-in Sensitive Information Types (SITs) and trainable classifiers are not automatically applied to Fabric assets during the metadata extraction phase.In practice, most organizations handle this in one of two ways: either they classify the data at the source system (for example in SQL or ADLS) where Purview scanning supports automatic classification, and then propagate that governance metadata into Fabric, or they apply classifications and sensitivity labels directly at the Fabric item level using Microsoft Purview Information Protection policies after the assets are discovered. Since Fabric scanning currently prioritizes metadata and lineage, classification often becomes a governance workflow step rather than an automatic scan outcome.
So for now, if classification based on SITs is required, it typically needs to happen upstream or through labeling policies, rather than being generated automatically during the Fabric metadata extraction process in Purview.
I hope this information helps. Please do let us know if you have any further queries.
Thank you
6 Replies
Hello fabricpribeiro,
for your need to classify Fabric assets based on sensitive information, or AI classifiers, you will need to use Microsoft Purview Information protection.
You can have auto labeling policies to meet some criteria and automatically labels Fabric items (lakehouses, etc...).
You can also check the video from Guy in a Cube to understand the licence part :
https://www.youtube.com/watch?v=nernTDBovtw
Hope it can help you!
Best regards,
Antoine
Hello, thank a lot for your replies. In this case I am speaking about classification during extraction. Sensitivty labels is another step. What I need to understand is how people are classififying based on the fact that Purview, for the time being, does not allow it for Fabric (as opposite to other sources like databricks or SQL)
Hi @fabricpribeiro ,
Thanks for reaching out to the Microsoft Fabric Community forum.
At the moment, Purview’s Data Map scans for Microsoft Fabric do not yet perform content-based classification during the scan the way they do for some other sources like SQL or Databricks. Because of this limitation, the built-in Sensitive Information Types (SITs) and trainable classifiers are not automatically applied to Fabric assets during the metadata extraction phase.In practice, most organizations handle this in one of two ways: either they classify the data at the source system (for example in SQL or ADLS) where Purview scanning supports automatic classification, and then propagate that governance metadata into Fabric, or they apply classifications and sensitivity labels directly at the Fabric item level using Microsoft Purview Information Protection policies after the assets are discovered. Since Fabric scanning currently prioritizes metadata and lineage, classification often becomes a governance workflow step rather than an automatic scan outcome.
So for now, if classification based on SITs is required, it typically needs to happen upstream or through labeling policies, rather than being generated automatically during the Fabric metadata extraction process in Purview.
I hope this information helps. Please do let us know if you have any further queries.
Thank you
Hi fabricpribeiro ,
Thanks for reaching out to the Microsoft Fabric Community forum.
Microsoft Purview’s Data Map works by scanning registered data sources, such as Microsoft Fabric, to automatically extract and store their metadata (for example tables, columns, schemas, and relationships) in a centralized metadata repository. The Data Map itself focuses mainly on metadata discovery and relationship mapping, This map is kept up to date by its integrated scanning and classification system, and all Microsoft Purview accounts have a data map that starts with a unit of capacity and can grow elastically.
Please refer to these articles to learn how the data scan and classification works:
Data classification in Microsoft Purview Data Map | Microsoft Learn
Automatically Apply Classifications on Data Map Assets | Microsoft Learn
Scans and ingestion in Microsoft Purview Data Map | Microsoft Learn
System classifications supported in Microsoft Purview Data Map | Microsoft Learn
I hope this information helps. Please do let us know if you have any further queries.
Thank youThank you very much for your reply
About this part " or they apply classifications and sensitivity labels directly at the Fabric item level using Microsoft Purview Information Protection policies after the assets are discovered"
I think Purview is not able to automatically apply labels to fabric items, not even via auto-labelling
If I am not wrong the only thing we can do is , using DLP , ask purview to check if there are any sensitive information and if so, to alert us, so that we can go manually to the fabric item and apply the label
Is it correct?
Thanks a lot,
Pedro
P.S - about "Purview scanning supports automatic classification, and then propagate that governance metadata into Fabric" how do you propaga it back to Fabric?
Hi fabricpribeiro ,
You’re mostly correct, but there’s a small nuance worth clarifying.
DLP policies, however, do work as expected. They can detect sensitive data and trigger alerts or policy actions. But as you mentioned, this is primarily a detection and monitoring mechanism, not an automatic labeling mechanism.
In other words, the current model for Fabric is more governance-driven (metadata + processes) rather than fully automated classification and labeling.
Regards
