This document describes how to configure sensitive data discovery for
Vertex AI data at the project level. For instructions on how to
configure discovery at the organization or folder level, see [Profile
Vertex AI data in an organization or
folder](https://docs.cloud.google.com/sensitive-data-protection/docs/profile-org-folder-vertex-ai).


For more information about the discovery service, see [Data
profiles](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles).

## Before you begin

1. If you have an organization-level discovery subscription---including one
   through Security Command Center---be aware that this project-level discovery
   configuration isn't included in your subscription and is billed separately.
   We recommend that you use an organization-level discovery configuration to
   profile the project. For more information, see [Profile select projects or
   data assets in an organization or
   folder](https://docs.cloud.google.com/sensitive-data-protection/docs/common-discovery-configurations#profile-select-projects-assets).

2. Make sure the Cloud Data Loss Prevention API is enabled on your project:

3. Confirm that you have the IAM permissions that
   are required to configure data profiles at the [project](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles.md.txt#permissions-to-configure-project) level.

4.
   You must have an inspection template in each region where you have data to be
   profiled. If you want to use a single template for multiple regions, you can use
   a template that is stored in the `global` region. If organizational
   policies prevent you from creating an inspection template in the `global` region, then
   you must set a dedicated inspection template for each region. For more
   information, see [Data
   residency considerations](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles#data-residency).

   This task lets you create an inspection template in the `global` region only.
   If you need dedicated inspection templates for one or more regions, you must
   [create those
   templates](https://docs.cloud.google.com/sensitive-data-protection/docs/creating-templates-inspect#create_a_new_inspection_template)
   before performing this task.
5. You can configure Sensitive Data Protection to send notifications to
   Pub/Sub when certain events occur, such as when
   Sensitive Data Protection profiles a new resource. If you want to use this
   feature, you must first [create a Pub/Sub
   topic](https://docs.cloud.google.com/pubsub/docs/admin#create_a_topic).


## Create a scan configuration

1. Go to the **Create scan configuration** page.

   [Go to Create scan configuration](https://console.cloud.google.com/security/sensitive-data-protection/create/discoveryConfiguration;source=7)
2. Go to your project. On the toolbar, click the project selector and
   select your project.

The following sections provide more information about the steps in the **Create
scan configuration** page. At the end of each section, click **Continue**.

### Select a discovery type

Select **Vertex AI**.

### Select scope

Do one of the following:

<br />

- If you want to scan a single resource, select **Scan one
  resource**.

  For each resource, you can have only one single-resource scan
  configuration. For more information, see [Profile a single data
  resource](https://docs.cloud.google.com/sensitive-data-protection/docs/discovery-test-mode).

  Fill in the details of the resource that you want to profile.
- If you want to perform standard project-level profiling, select **Scan
  selected project**.

### Manage schedules

If the [default profiling
frequency](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles#scan-frequency) suits
your needs, you can skip this section of the **Create scan configuration** page.

Configure this section for the following reasons:

- To make fine-grained adjustments to the profiling frequency of all your data or certain subsets of your data.
- To specify the resources that you don't want to profile.
- To specify the resources that you don't want profiled more than once.

To make fine-grained adjustments to profiling frequency, follow these steps:

1. Click **Add schedule**.
2. In the **Filters** section, you define one or more filters that specify
   which resources are in the schedule's scope. A resource is
   considered to be in the schedule's scope if it matches at least one of the
   filters defined.

   To configure a filter, specify a project ID or a regular expression that
   specifies one or more projects.

   Regular expressions must follow
   [RE2 syntax](https://github.com/google/re2/wiki/Syntax).

   For example, if you want all resources in a project to be included in
   the filter, enter the project ID in the **Project ID** field.

   If you want to add more filters, click **Add filter** and repeat this step.
3. Click **Frequency**.

4. In the **Frequency** section, specify whether the discovery service
   profiles the resources that you selected and, if so, how often:

   - If you never want the resources to be profiled, turn off
     **Do profile this data**.

   - If you want the resources to be profiled at least once, leave
     **Do profile this data** on.

     In the succeeding fields in this section, you specify whether the system
     should reprofile your data and what events should trigger a reprofile
     operation. For more information, see [Frequency of data profile
     generation](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles#scan-frequency).
     1. For **On a schedule**, specify how often you want the the resources to be reprofiled. The resources are reprofiled regardless of whether they underwent any changes.
     2. For **When inspect template changes** , specify whether you want your data to be reprofiled when the associated inspection template is updated, and if so, how often.

        > [!IMPORTANT]
        > **Note:** You specify the inspection templates to use in the [Select inspection template](https://docs.cloud.google.com/sensitive-data-protection/docs/profile-project-vertex-ai.md.txt#select-inspection-template) step on this page.

        An inspection template change is detected when either of the following occurs:
        - The name of an inspection template changes in your scan configuration.
        - The [`updateTime`](https://docs.cloud.google.com/sensitive-data-protection/docs/reference/rest/v2/projects.inspectTemplates.md.txt#InspectTemplate.FIELDS.update_time) of an inspection template changes.

        <br />

     3. For example, if you set an inspection template for the `us-west1` region and you update that inspection template, then only data in the `us-west1` region will be reprofiled.
5. Optional: Click **Conditions**.

   In the **Conditions** section, you specify any conditions that the
   resources---defined in your filters---must meet before
   Sensitive Data Protection profiles them.

   If needed, set the following:
   - **Time condition**: If you don't want old resources to ever be
     profiled, turn on this option. Then, use the date picker to select a date
     and time. Any resource created on or before your selected timestamp
     is excluded from profiling.

   - **Minimum condition**: If you want to delay profiling of a resource
     until it reaches a certain age, turn on this option. Then, enter the
     minimum duration.

   - **Resource types** : To limit profiling to certain Vertex AI
     resource types, clear the **All resource types** checkbox, and select the
     resource types that Sensitive Data Protection must profile.

     > [!IMPORTANT]
     > **Important:** If you skip this step, all Vertex AI resource types are profiled. If new Vertex AI resource types are later supported, those are automatically included in the configuration and are profiled as well.

   #### Example conditions

   Suppose that you have the following configuration:
   - **Time condition**

     - Timestamp: 05/4/22, 11:59 PM
   - **Minimum conditions**

     - Minimum duration: 24 hours
   - **Resource types**

     - Tuning job

   In this case, Sensitive Data Protection excludes any tuning job that was
   created on or before 11:59 PM on May 4, 2022. Among the tuning jobs that were
   created after that date and time, Sensitive Data Protection profiles only
   the tuning jobs that are at least 24 hours old. Other Vertex AI
   resource types aren't profiled.
6. Click **Done**.

7. Optional: To add more schedules, click **Add schedule** and repeat the
   previous steps.

8. To specify precedence between schedules, reorder them using the up and down arrows.

   The order of the schedules specifies how conflicts between schedules are
   resolved. If a resource matches the filters of two different schedules,
   the schedule higher in the schedules list dictates the profiling frequency
   for that resource.

   > [!NOTE]
   > **Note:** If your discovery pricing mode is [subscription
   > mode](https://cloud.google.com/sensitive-data-protection/pricing#discovery-subscription-mode-pricing), the rate at which Sensitive Data Protection profiles your data is affected by how much capacity you purchased. To determine your daily profiling capacity, see [Monitoring
   > utilization](https://cloud.google.com/sensitive-data-protection/pricing#subscription-monitoring). If you have [under-provisioned
   > capacity](https://cloud.google.com/sensitive-data-protection/pricing#underprovisioned), then the profiling frequencies that you set in your schedules might not be followed. If there is a backlog of data to be profiled, the schedule order doesn't dictate the order in which Sensitive Data Protection profiles the data in the backlog. Rather, all data resources in scope get a randomly assigned slot in the queue.

9. Optional: Edit or turn off **Catch-all schedule**.

   The last schedule in the list is the catch-all schedule. This schedule covers
   the resources in your selected scope that don't match any of the
   schedules that you created. The catch-all schedule follows the [system
   default profiling
   frequency](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles#scan-frequency).
   - To adjust the catch-all schedule, click **Edit schedule**, and then adjust the settings as needed.
   - To prevent Sensitive Data Protection from profiling any resource that is covered by the catch-all schedule, turn off **Profile the resources
     that don't match any custom schedule**.

### Select inspection template

Depending on how you want to provide an inspection configuration, choose one of
the following options. Regardless of which option you choose,
Sensitive Data Protection scans your data in the region where that data is stored.
That is, your data doesn't leave its region of origin.

#### Option 1: Create an inspection template

Choose this option if you want to create a new inspection template in the
`global` region.

1. Click **Create new inspection template**.
2. Optional: To modify the default selection of infoTypes, click **Manage infoTypes**.

   For more information about how to manage built-in and custom infoTypes, see
   [Manage infoTypes through the
   Google Cloud console](https://docs.cloud.google.com/sensitive-data-protection/docs/manage-infotypes-console).

   You must have at least one infoType selected to continue.
3. Optional: Configure the inspection template further by adding rulesets
   and setting a confidence threshold. For more information, see
   [Configure detection](https://docs.cloud.google.com/sensitive-data-protection/docs/creating-templates-inspect.md.txt#configure-detection).

When Sensitive Data Protection creates the scan configuration, it stores this
new inspection template in the `global` region.

#### Option 2: Use an existing inspection template

Choose this option if you have existing inspection templates that you
want to use.

1. Click **Select existing inspection template**.
2. Enter the full resource name of the inspection template that you want to use. The **Region** field is automatically populated with the name of the region where your inspection template is stored.

   The inspection template that you enter must be in the same region as the data
   to be profiled.

   To respect data residency, Sensitive Data Protection doesn't
   use an inspection template outside the region where that template is stored.

   To find the full resource name of an inspection template, follow these
   steps:
   1. Go to your inspection templates list. This page opens on a separate tab.


      [Go to inspection templates](https://console.cloud.google.com/security/sensitive-data-protection/landing/configuration/templates)

      <br />

   2. Select the project that contains the inspection template that you want to use.
   3. Select **Configuration \> Templates \> Inspect**, and then click the template ID of the template that you want to use.
   4. On the page that opens, copy the full resource name of the template. The full resource name follows this format:

      ```
      projects/PROJECT_ID/locations/REGION/inspectTemplates/TEMPLATE_ID
      ```
   5. On the **Create scan configuration** page, in the **Template name** field, paste the full resource name of the template.
3. To add an inspection template for another region, click **Add inspection
   template** and enter the template's full resource name. Repeat this for each region where you have a dedicated inspection template.
4. Optional: Add an inspection template that's stored in the `global` region. Sensitive Data Protection automatically uses that template for data in regions where you don't have a dedicated inspection template.

> [!CAUTION]
> **Caution:** If you don't include an inspection template that's stored in the `global` region, Sensitive Data Protection can't profile data in regions that don't have a dedicated inspection template. For more information, see [Data residency considerations](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles.md.txt#data-residency).

### Add actions

This section describes how to specify actions that you want
Sensitive Data Protection to take after profiling a resource. These actions
are useful if you want to send insights gathered from data profiles to other
Google Cloud services.

> [!NOTE]
> **Note:** For information about how other Google Cloud services may charge you for configuring actions, see [Pricing for exporting data
> profiles](https://cloud.google.com/sensitive-data-protection/pricing#exporting-profiles-pricing).

<br />

#### Publish to Security Command Center

<br />

Findings from data profiles provide context when you triage and develop response
plans for your vulnerability and threat findings in
[Security Command Center](https://docs.cloud.google.com/security-command-center/docs.md.txt).

> [!NOTE]
> **Note:** You can also configure Security Command Center to automatically prioritize resources for the [attack path simulation
> feature](https://docs.cloud.google.com/security-command-center/docs/attack-exposure-learn) according to the calculated sensitivity of the data that the resources contain. For more information, see [Set resource priority values automatically by data
> sensitivity](https://docs.cloud.google.com/security-command-center/docs/attack-exposure-learn#priority-values-auto-by-data-sensitivity).

Before you can use this action, Security Command Center must be activated at the organization level. Turning on Security Command Center at the organization level enables the flow of findings from integrated services like Sensitive Data Protection. Sensitive Data Protection works with Security Command Center in all service tiers.

<br />

If Security Command Center isn't activated at the organization level,
Sensitive Data Protection findings won't appear in
Security Command Center. For more information, see [Check the activation level of
Security Command Center](https://docs.cloud.google.com/security-command-center/docs/activate-scc-overview#check-scc-activation-level).


To send the results of your data profiles to Security Command Center, make sure the
**Publish to Security Command Center** option is turned on.

For more information, see [Publish data profiles to
Security Command Center](https://docs.cloud.google.com/sensitive-data-protection/docs/send-profiles-to-scc).

<br />

#### Save data profile copies to BigQuery

<br />

Sensitive Data Protection saves a copy of each generated data profile
in a BigQuery table. If you don't provide the details of your
preferred table, Sensitive Data Protection creates a dataset and table in the
project.
By default, the dataset is named `sensitive_data_protection_discovery` and
the table is named `discovery_profiles`.

> [!IMPORTANT]
> **Important:** The output table uses [`DataProfileBigQueryRowSchema`](https://docs.cloud.google.com/sensitive-data-protection/docs/reference/rest/v2/DataProfileBigQueryRowSchema.md.txt) as its schema. This schema can change as Sensitive Data Protection adds features. Make sure that your workflows can handle schema changes, for example, by ignoring unknown fields.

This action lets you keep a history of all of your generated profiles. This
history can be useful for creating audit reports and [visualizing data
profiles](https://docs.cloud.google.com/sensitive-data-protection/docs/analyze-data-profiles). You can also
load this information into other systems.

Also, this option lets you see all of your data profiles in a single view,
regardless of which region your data resides in. Although you can also [view the
data profiles through the
Google Cloud console](https://docs.cloud.google.com/sensitive-data-protection/docs/manage-data-profiles), the
console displays the profiles in only one region at a time.

When Sensitive Data Protection fails to profile a resource, it periodically
retries. To minimize noise in the exported data, Sensitive Data Protection
exports only the successfully generated profiles to BigQuery.

Sensitive Data Protection starts exporting profiles from the time you turn on
this option. Profiles that were generated before you turned on exporting aren't
saved to BigQuery.

> [!IMPORTANT]
> **Note:**The service agent associated with your project must have write access on the table where the profile copies will be saved.

For example queries that you can use when analyzing data profiles,
see [Analyze data profiles](https://docs.cloud.google.com/sensitive-data-protection/docs/analyze-data-profiles.md.txt).

<br />

#### Save sample discovery findings to BigQuery

<br />

Sensitive Data Protection can add sample findings to a
BigQuery table of your choice. Sample findings represent a subset
of all findings and might not represent all infoTypes that were discovered.
Normally, the system generates around 10 sample findings per resource, but
this number can vary for each discovery run.

Each finding includes the actual string (also called *quote*) that was detected
and its exact location.

This action is useful if you want to evaluate whether your [inspection
configuration](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles#inspection-template) is correctly
matching the type of information that you want to flag as sensitive. Using the
exported data profiles and the exported sample findings, you can run
queries to get more information about the specific items that were flagged, the
infoTypes they matched, their exact locations, their calculated sensitivity
levels, and other details.

> [!IMPORTANT]
> **Important:** The output table uses [`DataProfileFinding`](https://docs.cloud.google.com/sensitive-data-protection/docs/reference/rest/v2/DataProfileFinding.md.txt) as its schema. This schema can change as Sensitive Data Protection adds features. Make sure that your workflows can handle schema changes, for example, by ignoring unknown fields.

To save sample findings to a BigQuery table, follow these
steps:

1. Turn on **Save sample discovery findings to BigQuery**.

2. Enter the details of the BigQuery
   table where you want to save the sample findings.

   The table that you specify for this action must be different from the
   table used for the **Save data profile copies to BigQuery** action.
   - For **Project ID**, enter the ID of an existing project where you want
     to export the findings to.

   - For **Dataset ID**, enter the name of an existing dataset in the project.

   - For **Table ID**, enter the name of the BigQuery table where
     want to save the findings to. If this table doesn't exist,
     Sensitive Data Protection automatically creates it for you using the name
     that you provide.

> [!IMPORTANT]
> **Note:**The service agent associated with your project must have write access on the table.

For information about the contents of each finding that is saved in the
BigQuery table, see
[`DataProfileFinding`](https://docs.cloud.google.com/sensitive-data-protection/docs/reference/rest/v2/DataProfileFinding.md.txt).

<br />

#### Publish to Pub/Sub

<br />

Turning on **Publish to Pub/Sub** lets you take programmatic
actions based on profiling results. You can use Pub/Sub
notifications to develop a workflow for catching and remediating findings
with significant data risk or sensitivity.

To send notifications to a Pub/Sub topic, follow these steps:

1. Turn on **Publish to Pub/Sub**.

   A list of options appears. Each option describes an event that causes
   Sensitive Data Protection to send a notification to Pub/Sub.
2. Select the events that should trigger a Pub/Sub notification.

   If you select **Send a Pub/Sub notification each time a profile is updated** ,
   Sensitive Data Protection sends a notification when there's a change in the
   sensitivity level, data risk level, detected infoTypes, public access, and
   other important [metrics](https://docs.cloud.google.com/sensitive-data-protection/docs/metrics-reference.md.txt) in the
   profile.
3. For each event you select, follow these steps:

   1. Enter the name of the topic. The name must be in the following format:

          projects/PROJECT_ID/topics/TOPIC_ID

      Replace the following:
      - <var translate="no">PROJECT_ID</var>: the ID of the project associated with the Pub/Sub topic.
      - <var translate="no">TOPIC_ID</var>: the ID of the Pub/Sub topic.
   2. Specify whether to include the full resource profile in the
      notification, or just the full resource name of the resource that
      was profiled.

   3. Set the minimum data risk and sensitivity levels that must be met for
      Sensitive Data Protection to send a notification.

   4. Specify whether only one or both of the data risk and sensitivity
      conditions must be met. For example, if you choose `AND`, then
      both the data risk and the sensitivity conditions must be
      met before Sensitive Data Protection sends a notification.

> [!IMPORTANT]
> **Note:** The service agent associated with your project must have [publishing access](https://docs.cloud.google.com/pubsub/docs/access-control.md.txt#console) on the Pub/Sub topic. An example of a role that has publishing access is the Pub/Sub Publisher role (`roles/pubsub.publisher`). If there are configuration or permission issues with the Pub/Sub topic, Sensitive Data Protection retries sending the Pub/Sub notification for up to two weeks. After two weeks, the notification is discarded.

#### Send to Knowledge Catalog as aspects

<br />

This action lets you add
[Knowledge Catalog aspects](https://docs.cloud.google.com/dataplex/docs/enrich-entries-metadata#aspects)
to profiled resources based on insights from data profiles.
This action is only applied to new and updated profiles.
Existing profiles that aren't updated aren't sent to Knowledge Catalog.

When you enable this action, Sensitive Data Protection attaches the
`Sensitive Data Protection profile` aspect to the [Knowledge Catalog
entry](https://docs.cloud.google.com/dataplex/docs/ingest-custom-sources#entries) for each new or updated
resource that you profile. The generated aspects contain insights gathered
from the data profiles. You can then search your organization and projects for
entries with specific `Sensitive Data Protection profile` aspect values.

To send the data profiles to Knowledge Catalog, make sure that the
**Send to Dataplex Catalog as aspects** option is turned on.

For more information, see [Add Knowledge Catalog aspects based
on insights from data profiles](https://docs.cloud.google.com/sensitive-data-protection/docs/add-aspects).


### Set fallback processing locations for images

In general, Sensitive Data Protection processes your data in the location
where the data is stored. However, images can only be processed in a
multi-region or in the `global` region. If you set a fallback location, then
Sensitive Data Protection uses your fallback location to process images that
aren't in a multi-region or in the `global` region. If you skip this section,
then those images aren't processed.

To set fallback locations for image processing, select one or both of the
following:

- **Fall back to the multi-region** : If an image can't be processed in its original location, then the image is processed in the [multi-region](https://docs.cloud.google.com/sensitive-data-protection/docs/locations.md.txt#multi-regions) that corresponds to the image's original location. If the image's original location has no corresponding multi-region, then the image is skipped.
- **Fall back to global** : If an image can't be processed in its original location, then the image is processed in the `global` region.

If you select both options, Sensitive Data Protection chooses which location
to use as a fallback location.

### Set location to store configuration

Click the **Resource location** list, and select the region where you
want to store this scan configuration. All scan configurations that you
later create will also be stored in this location.

Where you choose to store your scan configuration doesn't affect the data to be
scanned. Your data
is scanned in the same region where that data is stored. For more information,
see [Data residency considerations](https://docs.cloud.google.com/sensitive-data-protection/docs/data-profiles.md.txt#data-residency).

> [!NOTE]
> **Note:** If you already have an existing scan configuration, you can't change the value set in this field. All scan configurations are stored in the same location. If you want to change the location of all your scan configurations, you must [delete](https://docs.cloud.google.com/sensitive-data-protection/docs/manage-scan-configurations.md.txt#delete-config) them, recreate them, and store them in the new location.

### Review and create

1. If you want to make sure that profiling doesn't start automatically after you create the scan configuration, select **Create scan in paused mode** .

   This option is useful in the following cases:
   - You opted to save data profiles to BigQuery and you want to make sure the service agent has write access to the BigQuery table where the data profile copies will be saved.
   - You opted to save sample discovery findings to BigQuery and you want to make sure that the service agent has write access to the BigQuery table where the sample findings will be saved.
   - You configured Pub/Sub notifications and you want to [grant publishing access](https://docs.cloud.google.com/pubsub/docs/access-control.md.txt#console) to the service agent.
2. Review your settings and click **Create** .

   Sensitive Data Protection creates the scan configuration and adds it to the
   discovery scan configurations list.

To view or manage your scan
configurations, see [Manage scan
configurations](https://docs.cloud.google.com/sensitive-data-protection/docs/manage-scan-configurations).

<br />

> [!IMPORTANT]
> **Note:** We regularly improve our detection algorithm. If we find that your organization or project would benefit from a new improvement that we implement, we might automatically regenerate your data profiles and redo the [actions in your scan
> configuration](https://docs.cloud.google.com/sensitive-data-protection/docs/profile-org-folder#manage-scan-outcome). You won't incur Sensitive Data Protection charges for this operation. However, because we will redo the actions, you might incur charges for your use of other Google Cloud services. For example, if you configured Sensitive Data Protection to save the data profiles to BigQuery, you might incur BigQuery charges.

<br />

## What's next

- Learn how to [manage data profiles](https://docs.cloud.google.com/sensitive-data-protection/docs/manage-data-profiles.md.txt).
- Learn how to [manage scan configurations](https://docs.cloud.google.com/sensitive-data-protection/docs/manage-scan-configurations.md.txt).
- Learn how to [receive and parse Pub/Sub messages published by the data profiler](https://docs.cloud.google.com/sensitive-data-protection/docs/dp-receive-pubsub-messages.md.txt).
- Learn how to [troubleshoot issues with data profiles](https://docs.cloud.google.com/sensitive-data-protection/docs/troubleshoot-data-profiles.md.txt).
- Look through the [data profiling limits](https://docs.cloud.google.com/sensitive-data-protection/limits.md.txt#data-profiling-limits).