# Managing Prompt Security

Prompt Security intercepts AI prompts before they leave the device, scans for sensitive content, and enforces your configured policy. Depending on your selections, Prompt Security can redact sensitive data before it reaches the AI tool or log the data for auditing purposes. Coverage spans web-based AI tools accessed through a browser extension and AI tools accessed through endpoint plugins, with reports appearing on a single console in Agentic Fabric.

**Web-Based AI Tools**

A Chrome and Edge browser extension intercepts prompts submitted to Anthropic Claude, ChatGPT, Deepseek, Gemini Enterprise, Google Gemini, and Perplexity. The extension sends prompts to a cloud-based classification service for ML-powered scanning, then enforces the decision inline. Based on your configured policies, sensitive content is automatically redacted before the prompt reaches the AI tool. If the cloud service is unreachable, the prompt is submitted, so developers are never interrupted by an outage. Deployment is silent via any Mobile Device Management (MDM) platform, as well as installation via a generic PowerShell script.

**Endpoint AI Tools**

Developers often work through IDE-integrated and CLI-based AI tools that never touch a browser. Endpoint plugins are installed through each tool’s marketplace or plugin configuration.

**Scanning Engine**

Web-based and Endpoint AI tools share the same scanning engine that detects:

- **Credentials**: API keys (AWS, OpenAI, Anthropic, GCP, Azure, GitHub, Stripe, and more), database connection strings, private keys (PEM/SSH), JWTs, OAuth/Bearer tokens, webhook signing secrets, and high-entropy secret strings.

- **Personally Identifiable Information (PII)**: Social Security numbers, ITIN, credit card numbers, email addresses, phone numbers, dates of birth, passport and national ID numbers, driver’s license numbers, IBAN/SWIFT, bank account and routing numbers, and IP/MAC/device identifiers.

- **Protected Health Information (PHI)**: Medical record numbers, ICD-10/CPT codes, prescription and Rx numbers, diagnosis and clinical details, lab results, and insurance member IDs.

The engine uses an ML-based classifier that evaluates prompts against tenant-configured policies and returns a decision (allow or redact) with explanations for each finding.

## Enabling Prompt Security

Prompt Security is disabled by default. Complete the following steps to enable it:

1. Go to **Admin > Global > System Settings**.
2. In the left pane, go to the **Feature Settings** section and select **Other Features**.
3. Scroll to the **Prompt Security** card.
4. Select the **Enable Prompt Security** toggle.
5. Scroll to the **Prompt Storage** card.
6. Select the **Allow Prompt Storage** toggle to allow prompts to be stored in a database and shown in the **Prompts** column of the **Activity Center** page.
7. Select **Save** to save your changes.

## Configuring the Sensors that Use Prompt Security

To make use of prompt security for web-based AI tools, configure the SailPoint Browser Agent Security browser extension sensor for visibility and support for GenAI governance processes. Refer to [Configuring the Browser Extension Sensor](../onboarding/sensors/browser_sensor/index.md) for more information.

To make use of prompt security for endpoint AI tools, configure SailPoint Endpoint Agent Security to discover and monitor AI agent software on managed laptops and desktops. Refer to [Configuring Endpoint Agents](../onboarding/sensors/endpoint_sensor.md) for more information.

## Configuring Prompt Security Policy

1. Go to **Agentic Fabric**.
2. From the left pane, go to the **Policy Center** section and select **Prompt Security Policy**.
3. In the **Response Timeout** field, enter the maximum wait time in seconds for cloud analysis. If this limit is reached, the prompt is allowed to prevent the user's application from freezing.
4. In the **Generative AI Business Applications** card, select the checkboxes to enable or disable scanning for each tool and tool type. The AI plugins and browser extensions support specific applications.
5. In the **Target Users and Group** card, define the target users and groups:

- Choose **All Users & Groups** to include all users and groups. You have the option to specify users and groups to exclude.
    - Choose **Specific Users & Groups** to identify specific users and groups both to include and to exclude.

6. Configure data protection rules and set a severity level (Critical, High, Medium, Low) and action (Disable, Audit Only, or Redact) for each category (Credentials, PHI, PII).
7. Select **Save** to save your policy.

Information about prompts will be displayed in Agentic Fabric's [Activity Center](../activity_center.md).

Managing Prompt Security

Prompt Security intercepts AI prompts before they leave the device, scans for sensitive content, and enforces your configured policy. Depending on your selections, Prompt Security can redact sensitive data before it reaches the AI tool or log the data for auditing purposes. Coverage spans web-based AI tools accessed through a browser extension and AI tools accessed through endpoint plugins, with reports appearing on a single console in Agentic Fabric.

Web-Based AI Tools

A Chrome and Edge browser extension intercepts prompts submitted to Anthropic Claude, ChatGPT, Deepseek, Gemini Enterprise, Google Gemini, and Perplexity. The extension sends prompts to a cloud-based classification service for ML-powered scanning, then enforces the decision inline. Based on your configured policies, sensitive content is automatically redacted before the prompt reaches the AI tool. If the cloud service is unreachable, the prompt is submitted, so developers are never interrupted by an outage. Deployment is silent via any Mobile Device Management (MDM) platform, as well as installation via a generic PowerShell script.

Endpoint AI Tools

Developers often work through IDE-integrated and CLI-based AI tools that never touch a browser. Endpoint plugins are installed through each tool’s marketplace or plugin configuration.

Scanning Engine

Web-based and Endpoint AI tools share the same scanning engine that detects:

Credentials: API keys (AWS, OpenAI, Anthropic, GCP, Azure, GitHub, Stripe, and more), database connection strings, private keys (PEM/SSH), JWTs, OAuth/Bearer tokens, webhook signing secrets, and high-entropy secret strings.
Personally Identifiable Information (PII): Social Security numbers, ITIN, credit card numbers, email addresses, phone numbers, dates of birth, passport and national ID numbers, driver’s license numbers, IBAN/SWIFT, bank account and routing numbers, and IP/MAC/device identifiers.
Protected Health Information (PHI): Medical record numbers, ICD-10/CPT codes, prescription and Rx numbers, diagnosis and clinical details, lab results, and insurance member IDs.

The engine uses an ML-based classifier that evaluates prompts against tenant-configured policies and returns a decision (allow or redact) with explanations for each finding.

Enabling Prompt Security

Prompt Security is disabled by default. Complete the following steps to enable it:

Go to Admin > Global > System Settings.
In the left pane, go to the Feature Settings section and select Other Features.
Scroll to the Prompt Security card.
Select the Enable Prompt Security toggle.
Scroll to the Prompt Storage card.
Select the Allow Prompt Storage toggle to allow prompts to be stored in a database and shown in the Prompts column of the Activity Center page.
Select Save to save your changes.

Configuring the Sensors that Use Prompt Security

To make use of prompt security for web-based AI tools, configure the SailPoint Browser Agent Security browser extension sensor for visibility and support for GenAI governance processes. Refer to Configuring the Browser Extension Sensor for more information.

To make use of prompt security for endpoint AI tools, configure SailPoint Endpoint Agent Security to discover and monitor AI agent software on managed laptops and desktops. Refer to Configuring Endpoint Agents for more information.

Configuring Prompt Security Policy

Go to Agentic Fabric.
From the left pane, go to the Policy Center section and select Prompt Security Policy.
In the Response Timeout field, enter the maximum wait time in seconds for cloud analysis. If this limit is reached, the prompt is allowed to prevent the user's application from freezing.
In the Generative AI Business Applications card, select the checkboxes to enable or disable scanning for each tool and tool type. The AI plugins and browser extensions support specific applications.
In the Target Users and Group card, define the target users and groups:
- Choose All Users & Groups to include all users and groups. You have the option to specify users and groups to exclude.
- Choose Specific Users & Groups to identify specific users and groups both to include and to exclude.
Configure data protection rules and set a severity level (Critical, High, Medium, Low) and action (Disable, Audit Only, or Redact) for each category (Credentials, PHI, PII).
Select Save to save your policy.

Information about prompts will be displayed in Agentic Fabric's Activity Center.

Documentation Feedback

Feedback is provided as an informational resource only and does not form part of SailPoint’s official product documentation. SailPoint does not warrant or make any guarantees about the feedback (including without limitation as to its accuracy, relevance, or reliability). All feedback is subject to the terms set forth at https://developer.sailpoint.com/discuss/tos.