Skip to main content
This page covers how the built-in Data Loss Prevention (DLP) guardrail works and how to enable it in AI Studio. After you enable it, Writer blocks prompts that contain selected categories of sensitive data before the model is invoked. There is no third-party provider to configure and no additional credentials to manage.
Only IT admins can enable and configure the DLP guardrail. Each organization can have one DLP guardrail.

​
What the guardrail detects

The DLP guardrail scans for the following categories of sensitive data. You can enable or disable individual categories when you configure the guardrail.
  • Personal information—names, email addresses, phone numbers, street addresses, and other personally identifiable information
  • Financial data—credit card numbers, bank account numbers, and related financial identifiers
  • Government IDs—national identification numbers, Social Security numbers, passport numbers, and similar government-issued identifiers
  • Healthcare identifiers—medical record numbers, health insurance identifiers, and other protected health information identifiers
  • Secrets and credentials—API keys, access tokens, passwords, and other credentials that could grant unauthorized access
When the guardrail detects sensitive data in an enabled category, it blocks the entire request. The model never receives the content. The guardrail does not redact or mask sensitive content and continue.

​
How the guardrail runs

The DLP guardrail runs a pre-call check on prompt text and files uploaded to a session. When it blocks a request, later guardrails in the execution order do not run. Execution order follows each guardrail’s priority number. If multiple guardrails share the same priority, the one created earlier runs first. You can configure DLP to fail closed or fail open when the DLP service is unavailable:
  • Fail closed—block all requests until the service recovers
  • Fail open—allow requests through until the service recovers
You can also scope the guardrail to specific teams and models. If you select both specific teams and specific models, the guardrail applies only where they overlap—not to all selected teams or all selected models independently.
The current release does not inspect model output (post-call), data retrieved through connectors, tool calls, or Knowledge Graph. Post-call detection is planned for a later release.

​
Enable the DLP guardrail

IT admins enable the DLP guardrail in AI Studio so it applies across the teams and models you select.
1

Open Guardrails in AI Studio

Log in to AI Studio, then go to Models & guardrails > Guardrails. Find the Writer-provided Data Loss Prevention guardrail and turn it on to open the setup modal.The modal opens on Basic info. Mode is Pre-call only. Set a priority. Lower numbers run first. If two guardrails share a priority, the one created earlier runs first. Select Next.
2

Select categories to block

On Detection, choose the categories of sensitive data to block, or accept the default set: Personal information, Financial, Government IDs, Healthcare, and Secrets & credentials. You can change these later.
3

Choose fail behavior

Under If the guardrail is unable to run, choose what happens if the detection service is unavailable:
  • Block the request (fail closed)—block all requests until the service recovers
  • Allow the request (fail open)—allow requests through until the service recovers
4

Scope teams and models

Under Apply to and Apply to models, select all teams and models, or limit the guardrail to specific ones. If you select both specific teams and specific models, the guardrail applies only where they overlap—not to all selected teams or all selected models independently. All models includes current and future models. Select Next.
5

Test and save

On Preview, enter sample input and select Preview. Confirm the guardrail blocks sensitive content, then select Save changes.
If you have multiple guardrails enabled (for example Bedrock and Azure), the one with the lower priority number runs first. If multiple share the same priority, the one created earlier runs first. When DLP blocks a request, later guardrails do not run and the model is not invoked.
Start by blocking only the most sensitive categories, test on the next screen, and consider scoping the guardrail to one team before rolling it out more widely.
You can disable the guardrail at any time by turning the toggle off on the Guardrails page.

​
How users experience a blocked request

When a user submits a prompt that contains sensitive data matching an enabled category, the guardrail blocks the request before the model call. The user receives a message indicating that the request was blocked to protect sensitive information, and no content is sent to the model. The original session remains intact. The user can revise the prompt to remove the sensitive data and resubmit, or continue working in the same session.

​
Monitor guardrail decisions

Allow and block decisions, including which categories triggered a block, appear in event logs and traces. You can also filter for them with the guardrail_outcome parameter on the Logs Export API.

​
Next steps

After you enable DLP, review related guardrail and logging guides.
  • Configure guardrails: Learn about guardrail modes, team scoping, and provider integrations
  • Event logs: View guardrail allow and block decisions in AI Studio
  • Export logs: Stream logs and filter by guardrail_outcome