What you will be able to do
- Explain how the system.certification_status tag tells authoritative assets apart from derived or outdated ones
- Apply certified and deprecated status with SQL and know which permissions this requires
- Plan certification at scale and check it with workspace search before an agent is deployed
Key concept
Trust signal vs. access boundary — Certifying an asset records that your organization vouches for it, and assistants such as Genie One use that to rank answers. Certification does not stop anyone from reading an asset. Only Unity Catalog privileges decide what an agent can actually reach.
1.Why an agent needs to know which assets are authoritative
In a mature Unity Catalog metastore, the tables that define your business sit next to many assets built from them: summaries, scratch copies, old snapshots and experiments. A person browsing the catalog can usually tell them apart from names and context. An agent that retrieves context cannot. To the agent, main.sales.transactions and main.sales.old_summary are both readable tables that look relevant. So before deployment, the governance strategy has two jobs. First, record which assets are authoritative. Second, make sure the agent's reach matches that record. This page covers the first job.
Unity Catalog records authority with certification. Domains, metric views and Pages describe what data means. Certification sits on top of those and says whether the organization stands behind a particular asset. Databricks describes it this way: "certification adds a layer of trust and authority on top of those definitions". This signal is what Genie One uses to prioritize assets when it answers questions.
Checkpoint 1 of 5· Check yourself
Domains, metric views and Pages define what your data means. What does certification add on top of them?
Certification does not define meaning. It adds a trust and authority signal on top of the definitions that domains, metric views and Pages provide.
“certification adds a layer of trust and authority on top of those definitions”Source: docs.databricks.com
Sources1
2.The system.certification_status tag
The tag key is system.certification_status. Databricks defines it as a system tag, which is a special kind of governed tag. Because Databricks owns the definition, every workspace in the account uses the same two values. Users cannot rename the key or add a third value. The tag can go on catalogs, schemas, tables, views, volumes, functions, registered models, dashboards, Genie Agents, Databricks Apps and notebooks. That covers both the data an agent retrieves and some of the AI assets that consume it.
| Value | Meaning | Shown in the workspace as |
|---|---|---|
| certified | The asset has met internal standards for accuracy, completeness, and trust | A check mark |
| deprecated | The asset is outdated, no longer reliable, or should not be used in new workflows | A restricted icon |
-- Apply certified tag key
SET TAG ON TABLE main.sales.transactions `system.certification_status` = `certified`;In the workspace UI, you do the same thing from the object's kebab menu with Assign certification. You can change or remove the status at any time. For the retrieval strategy, deprecation matters as much as certification. Marking a derived summary as deprecated tells every consumer, human or agent, that it should not be used in new workflows.
Checkpoint 2 of 5· Fill the gap
Which value marks the old derived summary so consumers avoid it?
SET TAG ON TABLE main.sales.old_summary `system.certification_status` = ` ? `;The tag accepts only certified or deprecated. Deprecated is the value for outdated assets. In the UI, 'None' removes the status rather than being a stored value.
Source: docs.databricks.comSources1
3.Who is allowed to certify
A trust signal is only worth something if not everyone can set it. If any table owner could certify their own derived table, the label would mean nothing. For this reason, applying system.certification_status needs two separate permissions. First, the normal tagging rights on the object: you own it, or you have APPLY TAG on the object plus USE SCHEMA and USE CATALOG on its parents. Second, the ASSIGN permission on the governed tag itself. Admins control ASSIGN through governed tag permission settings. That lets you limit certification to a data stewardship group.
Checkpoint 3 of 5· Check yourself
A data engineer owns a derived table and wants to mark it certified so the agent prefers it. They have no permissions on the certification tag. What happens?
Ownership covers ordinary tagging. Certification is a governed system tag, so it also needs the ASSIGN permission on that tag.
“You must have the ASSIGN permission on the system.certification_status governed tag to apply it to objects.”Source: docs.databricks.com
Sources1
4.Certifying at scale and checking the result
Tagging one table at a time does not scale to a large catalog. Databricks recommends automating tag assignment (beta). An automation applies system.certification_status to every Unity Catalog table or volume that matches conditions you set, such as query usage, asset age, owner, description, name or existing tags. Its rules are deterministic and are evaluated against metadata, not against the data itself. The documented example targets derived clutter directly: it deprecates every table in the temp catalog that has had no read or write queries in the last 30 days, then emails the owners.
Before deployment, check the result in workspace search. The certificationStatus keyword filters objects by status. You can list the certified tables you expect the agent to rely on, and you can search certificationStatus:deprecated to find assets that must stay out of its scope. Keep two limits in mind. Tag search does not work on dashboards, Genie Agents or Databricks apps. And updates can take a few minutes to show up in search. For a query-based audit, each catalog's INFORMATION_SCHEMA.TABLE_TAGS lists the tags on its tables.
type:table certificationStatus:certifiedCheckpoint 4 of 5· Check yourself
You want to check certification on a Genie Agent before deployment. Why might a certificationStatus search miss it?
Genie Agents can be certified, but tag-based search does not cover them. You need to check them another way, for example on the object itself.
“Search using tags is not supported on dashboards, Genie Agents, or Databricks apps.”Source: docs.databricks.com
Checkpoint 5 of 5· Exam question
A Databricks agent deployment needs to retrieve context only from an authoritative "gold" schema in Unity Catalog, while derived "silver" and "bronze" schemas containing intermediate or duplicated data must remain out of scope. The agent connects to AI Search using a dedicated service principal. Which governance action best constrains the agent's retrieval space before deployment?
Correct answer: A — Grant the agent's service principal SELECT privileges only on the gold schema, and do not grant SELECT on the silver or bronze schemas
- A. Correct. Unity Catalog SELECT grants on the underlying identity are the actual enforcement mechanism for retrieval scope; withholding SELECT on silver and bronze makes those schemas inaccessible regardless of what the agent is instructed to do.
- B. Incorrect. Prompt instructions are advisory, not enforced. If the service principal still has SELECT on silver and bronze, a user or a generated query can still surface that data.
- C. Incorrect. Indexing derived schemas puts that data within retrieval reach; relying on the LLM to self-filter at generation time is not a governance control and can fail or be bypassed.
- D. Incorrect. A naming or folder convention has no effect on access; without a corresponding permission change, the agent can still read the derived schemas.
- E. Incorrect. A shorter query timeout affects performance, not access, and does nothing to prevent the agent from reading silver or bronze data within that timeout.
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Owning a table is enough to mark it certified.Why is that wrong?
Ownership covers ordinary tags. The certification tag is governed, so you also need the ASSIGN permission on it.
Covered in Who is allowed to certify
2.Admins can add custom values to system.certification_status, such as a 'gold' tier.Why is that wrong?
Databricks predefines system tags, and users cannot modify their keys or values.
Covered in The system.certification_status tag
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.https://docs.databricks.com/aws/en/data-governance/unity-catalog/certify-deprecate-dataOfficial docs
“certification adds a layer of trust and authority on top of those definitions”
↩︎ Why an agent needs to know which assets are authoritative“deprecated: Warns that a data asset is outdated, no longer reliable, or should not be used in new workflows.”
↩︎ The system.certification_status tag“certified: Indicates that a data asset has met internal standards for accuracy, completeness, and trust.”
↩︎ The system.certification_status tag“You must have the ASSIGN permission on the system.certification_status governed tag to apply it to objects.”
↩︎ Who is allowed to certify“It matches every table in the temp catalog with no read or write queries in the last 30 days”
↩︎ Certifying at scale and checking the result“It might take a few minutes for certification or deprecation updates to appear in search results.”
↩︎ Certifying at scale and checking the result“This helps Genie One improve its answers by prioritizing the assets your organization vouches for.”
↩︎ Key concept“It is a system-governed tag with two tag values: certified and deprecated.”
↩︎ Prediction“Search using tags is not supported on dashboards, Genie Agents, or Databricks apps.”
↩︎ Checkpoint - 2.
“Deterministic rules, evaluated against metadata”
↩︎ Certifying at scale and checking the result
Also cited
“To add a governed tag to Unity Catalog securable objects, you must also have the ASSIGN permission on the governed tag.”
↩︎ Exam trap 1“Users cannot modify or delete system tag keys or values.”
↩︎ Exam trap 2