Supported editions for this feature: Frontline Standard and Frontline Plus, Enterprise Standard and Enterprise Plus, Education Fundamentals, Education Standard, and Education Plus; Enterprise Essentials Plus. Compare your edition
Using data loss prevention (DLP) for Google Workspace Studio, you can protect data in Workspace Studio flows. You create data protection rules that control flow activity and mitigate security risks that extract or manipulate sensitive data. You can scan each step's data sources, the content entered by a user to set up the step, and the people who get the results. For an activity that's sensitive or shared beyond data boundaries, you can block that step and cancel the run. Or, you can require user approval before the flow continues.
How does DLP for Studio work?
When a flow runs, DLP for Studio checks for sensitive data and data boundaries for each step in the flow. If the rule conditions are met, the rule action (block, require approval, or audit) is applied before the step runs. You can set up DLP for Studio to check:
- The sources of variables used in a step, such as if a file is classified as need-to-know.
- The user's setup of each step, such as if text in prompts or names for items contain sensitive data.
- The visibility of the step's outputs, such as if the step will send a message to an external user.
DLP for Studio in practice
- You define data protection rules. These rules define which sources and actions are sensitive and should be protected.
- DLP scans content for rule violations that trigger DLP incidents.
- If the rule is violated, actions are triggered.
- You are alerted to data protection rule violations.
DLP tries to scan all content and actions for each run of flows created by users in your organization. If your rule requires a user to approve an action, DLP will continue scanning the remaining steps when the user approves continuing the run.
Supported content types
DLP for Studio supports data from Google Drive, Chat, and Calendar, Gmail, and third-party apps.
Known limitations
DLP for Studio has the same extracted text-analysis limits and general rule and detector limitations as other Workspace Studio services.
For example, for a Workspace Studio step, only the first 10 MB of extracted text is analyzed.
For details, go to View DLP content and rule size limits.
DLP for Studio doesn't check resources the user linked to in AI-powered steps.
AI-generated text from sources with classification labels doesn't inherit the labels and aren't considered labeled by Workspace Studio.
DLP for Studio might behave differently than DLP for Gmail. DLP for Studio determines if a recipient is external based on whether the recipient belongs to the same Google Workspace Studio organization (using the same customer ID). DLP for Gmail might use more specific Gmail settings, such as trusted domains or domain aliases, to define what is internal or external.
If Drive shares a file with a group in Google Groups that contains external members, DLP for Studio might not recognize the file as external.
DLP for Studio won't scan files owned by a user outside of your organization.
DLP for Studio doesn't inspect the content or classification labels of any Drive files or Gmail messages linked using URLs.
When resources are linked in other Workspace Studio services, DLP for Studio evaluates the resource as part of the linking service, not the original service. For example, if a Drive file is linked within a Calendar event, DLP for Studio evaluates the Drive file as part of Calendar and not Drive.
You can't add custom messages for the Require user approval action.
DLP for Studio won't check policies when the user stars or unstars an email, marks an email as read or unread, or archives an email.
Understand DLP actions
When sensitive content is found or a data boundary could be crossed, your rule can enforce one of the actions included in the following table.
If you have similar rules with different response actions, the stricter action prevails. For example, if one rule only logs an event when a Social Security Number (SSN) is found and another rule blocks the flow from using SSNs, the flow can't continue the run because SSNs are blocked.
| Action | Description |
|---|---|
| Block Studio flows |
Stops the current run of the flow and notifies the user on the Workspace Studio Activity page. Optionally, you can add a custom message for users. The event is logged. This action doesn't block other instances of the flow from running. For example, a user might have a flow that researches important emails and notifies them in Chat. If your data protection rule blocks sharing of emails labeled confidential, it will only block runs that start from confidential emails. Runs that start from non-confidential emails won't be affected. |
| Require user approval |
Pauses the run and requires the user to review and approve the step on the Workspace Studio Approvals page before the run can continue. The user's choice is recorded in the log events. Rules are re-evaluated when a user approves and enforced if agent access is blocked. If the re-evaluation requires another user approval or is audit only, the rule isn't enforced because the user has already approved. |
| Audit only |
Lets the flow proceed without interruption and logs the event. |
Understand DLP content conditions
You can create a data protection rule with no conditions. In this case, the rule applies to everything in the flow.
Or, you can specify conditions in the data protection rule that define what content or activity to scan for. You can use predefined data types or create your own custom content detectors. You can also combine multiple conditions using AND, OR, or NOT operators.
For details, go to How to use predefined content detectors, Create a custom detector, and Examples of rules with nested condition operators.
Available content conditions
| Content type to scan | What to scan for | Details & use |
|---|---|---|
| Referenced sources |
Matches predefined data type Contains text string Matches classification label (Gmail and Drive only) Matches label Matches regular expression Matches words from list Any data |
Scans the Workspace Studio data sources for variables used by the step, including their content and metadata, such as classification labels. However, for AI-powered steps, Workspace Studio data sources that Gemini finds aren't scanned. This option can be overly-restrictive because it doesn't assess if the sensitive content is actually used by the step. For example, if any data in an email triggers a data protection rule, then using just the sender's email address causes a match. |
| Step inputs |
Matches predefined data type Contains text string Matches regular expression Matches words from list Any data |
Scans the content used to set up the step, content in variables, and any text or other direct user inputs, such as text strings and prompts. This option doesn't scan resources given to Gemini as links in prompts or Workspace Studio sources Gemini finds on its own. |
Understand DLP audience conditions
When you create a data protection rule, you can specify conditions that define who can see the output of the agent or the flow:
- Block sharing with external users and agents—Prevents Studio from sharing sensitive data outside your organization, but allows internal sharing.
- Block sharing with all users and agents—Prevents flows from all sharing, even with the flow creator.
Create a rule
After you decide what you want your rule to do, you create the rule. For details, go to Create data protection rules.
Common use cases
The following table gives examples of how to combine conditions (what is checked) and a specific action (the enforcement) to define your rule. To use this table, you must:
- Map condition values to the corresponding options.
- Select an action.
| Use case | Content condition | Audience condition | Action |
|---|---|---|---|
| Block external emails if need-to-know (NTK) content is found in the source email |
Referenced sources: Source data What to scan: Gmail Matches: Matches label Label: Sensitivity Label value: NTK |
Block sharing with external users and agents | Block Studio flows |
| Stop the flow if any data is shared externally |
Referenced sources: Source data What to scan: All apps Match: Any data OR Step attribute to scan: Step inputs What to scan: All apps Match: Any data |
Block sharing with external users and agents | Block Studio flows |
How data protection rules affect a flow
In the following example, a user creates a Workspace Studio flow to:
- Monitor a spreadsheet that's need-to-know for any changes.
- Ask Gemini to draft an email based on content from a variable that contains non-sensitive data from the row that changed in the sheet.
- Send the message generated by Gemini to a group that includes an external user.
| DLP rule conditions & action | Impact |
|---|---|
| If any Step inputs contain sensitive content, block the step. |
The run is successful because the data in the variable in step 2 isn't sensitive and the content generated by Gemini for step 3 isn't sensitive. |
| If Referenced sources OR Step inputs contain sensitive content, block the step. |
Run stops at step 2 because the source spreadsheet is sensitive, even though the data in the variable isn't. |
| If any Step inputs contain sensitive content OR audience conditions are set to external, block step. |
Run stops at step 3 because the step inputs aren't sensitive for any step, but that data will be shared externally. |
Review, monitor & investigate alerts
- View details of incidents in Rule log events and Workspace Studio log events.
- Investigate alerts of data-sharing incidents using the security investigation tool. For details, go to About the security investigation tool.
- Investigate data protection rule violations to determine if they're real incidents or false positives. For details, go to View content that triggers DLP rules.