Account Management and Data Governance · Topic 8 of 9

Data Classification and Governance in Snowsight

Detect sensitive columns with data classification, apply category tags, and manage and monitor governance and tags in the Snowsight governance pages.

  • 5 min
In a nutshell
  • Classification is Enterprise or higher and writes SNOWFLAKE.CORE.SEMANTIC_CATEGORY and SNOWFLAKE.CORE.PRIVACY_CATEGORY.
  • Privacy categories: IDENTIFIER, QUASI_IDENTIFIER, SENSITIVE.
  • EXTRACT_SEMANTIC_CATEGORIES(obj, max_rows_to_scan) analyzes with a warehouse; ASSOCIATE_SEMANTIC_CATEGORY_TAGS applies the top recommendation.

What it is

Data classification scans your tables and identifies columns that hold personal or sensitive data, such as names, emails and national identifiers. The result is recorded as system tags on the columns, so masking and other policies can act on it. It requires Enterprise Edition or higher.

Snowsight is where administrators manage governance without SQL: a Governance and security area with tags and policies, tag monitoring, and the Trust Center, which also holds the classification settings.

Table columnsscanned byClassificationsemantic and privacy categorywritesSystem tags on columnsused byMasking policyvia tag based masking

How it works

What classification produces

Each classified column gets two values.

ValueMeaningExamples
Semantic categoryWhat the data isName, email, national identifier
Privacy categoryHow sensitive it isIDENTIFIER, QUASI_IDENTIFIER, SENSITIVE

They are stored as the system tags SNOWFLAKE.CORE.SEMANTIC_CATEGORY and SNOWFLAKE.CORE.PRIVACY_CATEGORY. You can map your own tags to these, and tag-based masking policies then apply automatically.

The SQL workflow

The two classic steps are a function that analyzes and a procedure that applies.

-- Step 1: analyze (reads a sample, returns JSON)
SELECT EXTRACT_SEMANTIC_CATEGORIES('hr.public.employees');
SELECT EXTRACT_SEMANTIC_CATEGORIES('hr.public.employees', 5000);

-- Step 2: apply the reviewed result as tags
CALL ASSOCIATE_SEMANTIC_CATEGORY_TAGS(
  'hr.public.employees',
  EXTRACT_SEMANTIC_CATEGORIES('hr.public.employees'));

About EXTRACT_SEMANTIC_CATEGORIES:

  • The second argument is max_rows_to_scan, 1 to 10000, default 10000.
  • It needs a running warehouse and works on tables, external tables, views and materialized views.
  • The result is JSON with valid_value_ratio, a recommendation (semantic category, privacy category, confidence of HIGH, MEDIUM or LOW, and coverage) and a list of alternates.
  • NULLs, empty strings and strings over 256 characters are invalid values.

About the procedure, whose documented name is ASSOCIATE_SEMANTIC_CATEGORY_TAGS: it applies only the top level recommendation, not the alternates. To use an alternate, edit the stored result or set the tag by hand with ALTER TABLE ... MODIFY COLUMN ... SET TAG. To remove one, use UNSET TAG.

Both are marked legacy in the documentation, and Snowflake recommends the newer classification methods. They still work and are what the exam outline names.

Exam tip

The pattern is always extract, review, then associate. A reviewer checks the confidence and alternates before any tag is written.

Classification profiles and automatic classification

The newer method uses classification profiles. A profile holds the settings and controls which databases are classified, and associating the profile with a database starts classification. Settings are managed in the Trust Center in Snowsight and can also be created with SQL.

  • Autopilot (preview, the default for profiles) lets Snowflake manage timing, reclassification and minimum object age.
  • AI mode (preview) uses a model to find extra semantic categories and adds token charges to compute charges.
  • Classification uses serverless compute credits.
  • Supported objects include tables, external tables, Iceberg tables, dynamic tables, event tables and views. Unsupported data types include BINARY, GEOGRAPHY, UUID and VECTOR, and only JSON is supported among semi-structured types.
  • SYSTEM$SHOW_SENSITIVE_DATA_MONITORED_ENTITIES('DATABASE') lists the monitored databases.

Manage data governance in Snowsight

Snowsight groups governance tools under Governance and security: tags and policies, the Trust Center and related views. Requirements for the tags and policies area:

  • Enterprise Edition or higher.
  • The ACCOUNTADMIN role, or a role granted the GOVERNANCE_VIEWER and OBJECT_VIEWER database roles.
  • To create tags, the CREATE TAG privilege on the schema.

GOVERNANCE_VIEWER covers governance views such as ACCESS_HISTORY, TAG_REFERENCES and the masking and row access policy views; OBJECT_VIEWER covers object metadata. These are database roles on the SNOWFLAKE database, which is the safer way to give analysts account usage access than IMPORTED PRIVILEGES.

Configure and monitor tags in Snowsight

To create a tag: Governance and security, then Tags and policies, open the Tags tab and select + Create tag. Enter a name, database and schema and an optional description. Advanced options let you set allowed values and tag propagation, then save.

To monitor:

  • The dashboard shows tiles, distribution percentages and frequently used tags. Selecting an item opens the Tagged Objects tab.
  • On Tagged Objects you filter objects and manage tag assignments.
  • To see tags, you need USAGE on the schema that contains them, or the APPLY TAG privilege on the account.
  • The same facts are available in SQL with SYSTEM$GET_TAG, TAG_REFERENCES and TAG_REFERENCES_WITH_LINEAGE.
Governance and securityopenTags and policiescreate or editTagallowed values, propagationmonitor inTagged Objects

Common mistakes

Applying classification output without review. A low confidence result can tag a column wrongly, and tag-based masking would then mask the wrong data.

Expecting classification to mask data. It only tags. Masking happens when a policy is bound to those tags.

Classifying unsupported types. BINARY, GEOGRAPHY, UUID and VECTOR columns are not classified.

Giving analysts IMPORTED PRIVILEGES on the SNOWFLAKE database for governance. Use the narrower GOVERNANCE_VIEWER role instead.

Using Snowsight governance pages on a Standard account. They need Enterprise Edition or higher.

What to remember for the exam

  • Classification is Enterprise or higher and writes SNOWFLAKE.CORE.SEMANTIC_CATEGORY and SNOWFLAKE.CORE.PRIVACY_CATEGORY.
  • Privacy categories: IDENTIFIER, QUASI_IDENTIFIER, SENSITIVE.
  • EXTRACT_SEMANTIC_CATEGORIES(obj, max_rows_to_scan) analyzes with a warehouse; ASSOCIATE_SEMANTIC_CATEGORY_TAGS applies the top recommendation.
  • Classification profiles with autopilot and AI mode are the newer approach; both are preview features.
  • Snowsight tags need CREATE TAG; the governance area needs ACCOUNTADMIN or GOVERNANCE_VIEWER plus OBJECT_VIEWER.

Check yourself

  • What does EXTRACT_SEMANTIC_CATEGORIES return and what does it change?

    It returns JSON with a recommended category, confidence and alternates; it changes nothing until the result is applied as tags.

  • Which two system tags record classification results?

    SNOWFLAKE.CORE.SEMANTIC_CATEGORY and SNOWFLAKE.CORE.PRIVACY_CATEGORY.

  • Which database roles let a non-admin use the Snowsight tags and policies area?

    GOVERNANCE_VIEWER and OBJECT_VIEWER.

Topic 8 of 9 in Account Management and Data Governance

Test yourself on Account Management and Data Governance

The full Snowflake ADA-C02 question bank has 814 questions, with timed practice exams and an explanation for every answer.

This study guide is independent and is not affiliated with, endorsed by, or sponsored by Snowflake. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.