GenAI Prompt
Table of Contents
GenAI Prompt
GenAI Prompt controls
| Prisma Browser Desktop | Prisma Browser Extension | Prisma Browser for Mobile |
|---|---|---|
| Full support | Full support | No support |
| Content-aware prompt scanning is not supported. Allow and block enforcement only. |
GenAI Prompt Control enables administrators to monitor and control
data submitted to Generative AI applications and embedded AI widgets. The control
applies to standalone applications such as ChatGPT, Microsoft Copilot, Gemini,
Claude, and Perplexity, as well as AI widgets embedded within standard
business tools — for example, Gemini in Google Workspace applications such as Docs,
Drive, and Gmail, or Microsoft Copilot embedded in Microsoft Word and other
Microsoft 365 web applications.
The control operates across multiple enforcement layers, leveraging
last-mile analysis on the browser side to make it resilient against attempts to
bypass policy enforcement.
For selective enforcement, configure the Block
action together with the rule's When Contains section to
restrict blocking to prompts that match specific DLP profiles — for example,
blocking only prompts that contain PII, PHI, or proprietary content, while allowing
all other prompts through.
Comparison with Typing Guard
The GenAI Prompt control and the Typing Guard control both address data
entry in web applications, but they operate differently:
- Typing Guard (Block mode) - Masks sensitive data while still allowing the prompt to be submitted with the sensitive content replaced. The original prompt reaches the AI provider, minus the redacted portions.
- GenAI Prompt (Block) - Prevents the prompt from being submitted entirely. No part of the prompt reaches the AI provider.
Typing Guard operates at the DOM level only and may not reliably intercept
input in all AI application implementations. The GenAI Prompt control enforces at
multiple layers on the browser side, making it more suitable when complete
prevention of data submission is required.
Tenancy Enforcement
To restrict rule enforcement to specific enterprise tenants, configure the
Application Tenants step of the rule. This allows, for
example, blocking prompts only when the user is logged into the organization's
approved Claude organization, while allowing the same application when accessed with
a personal account.
Tenancy enforcement is available for the following providers:
| Provider | Identifier |
|---|---|
| Claude | Organization ID (UUID) or domain |
| ChatGPT (OpenAI) | Domain, tenant ID, plan type, or account structure |
| Microsoft (Copilot / M365) | Domain and resource host |
| Google (Workspace) | Domain |
Tenancy enforcement is configured in the
Application Tenants step of the Access and Data Control
rule, not in the GenAI Prompt control step itself.
Content Inspection
When set to Block, the Prisma Browser can
evaluate prompt text against configured DLP profiles before applying the enforcement
action. Two types of data profiles are supported:
- Local data profiles - Perform on-device pattern matching without sending prompt content to external services.
- Cloud-assisted data profiles - Send prompt text to Palo Alto Networks cloud scanning services, enabling advanced classifiers including Advanced Machine Learning, LLM-based classifiers, Exact Data Matching (EDM), and Indexed Data Matching (IDM).
Content inspection is configured using the When
Contains section of the rule.
Compliance Relevance
The GenAI Prompt control supports the following regulatory frameworks that
require visibility and governance over AI interactions:
- EU AI Act — Mandates tracking and auditing of AI system interactions for covered providers and deployers.
- GDPR / CCPA — Requires organizations to verify that PII is not transmitted to third-party AI vendors without appropriate controls.
- NIST AI RMF (AI 600-1) — Focuses on tracking harmful content, data provenance, and AI transparency.
- HIPAA — Requires controls to prevent PHI from being shared with AI vendors that are not covered entities or business associates.
Keep the following limitations in mind:
- GenAI prompt collection does not work with Safari.
- Gen AI prompt history collection does not work with PBX on Edge for Microsoft Copilot
To set the GenAI Prompt control:
- From Strata Cloud Manager, selectSelect GenAI Prompt.Select one of the following options:
- Allow - the Prisma Browser permits prompt submission. Prompt collection can be configured for allowed prompts in the rule's Tracking section.
- Block - the Prisma Browser prevents the prompt from being submitted entirely. To selectively block based on prompt content, configure the rule's When Contains section with DLP profiles. Optionally, configure a custom dialog message to display to the user when a prompt is blocked.
Click Set.![]()
