Prisma AIRS icon

Prisma AIRS

Prisma AIRS AI Security.

Overview

This node integrates with the Prisma AIRS AI Security service to inspect and protect AI chat interactions by analyzing either user prompts or AI-generated responses. It helps detect and block prompt injection attacks or unsafe AI outputs, enhancing security in conversational AI applications.

Common scenarios include:

  • Validating user inputs (prompts) before sending them to an AI model to prevent malicious injections.
  • Inspecting AI-generated responses to ensure they do not contain harmful or sensitive content.
  • Automatically blocking or masking detected threats in real-time chat systems.

Practical example:

  • In a chatbot application, before forwarding user input to the AI engine, this node checks the prompt for injection attacks using a configured Prisma AIRS profile. If an attack is detected, it blocks the input and returns a custom warning message.
  • After receiving the AI's response, the node inspects the output text for any policy violations or sensitive data leakage, masking or blocking the response as needed.

Properties

Name Meaning
Response (Required) The AI Agent's response text to be checked.
Prisma AIRS AI Profile for Output (Required) The Prisma AIRS AI profile name configured for output scanning.
Session ID (Optional) Unique identifier for the current chat session.
App User (Optional) User name associated with the current chat session.
User IP (Optional) IP address of the user in the current chat session.
App Name (Optional) Application name for the current chat session.
AI Model Name (Optional) AI model name used in the current chat session.
Custom Your Block Message Custom message returned when a prompt injection attack or unsafe response is detected.

Output

The node outputs JSON objects containing the inspection results with fields such as:

  • sessionId: The session identifier if provided.
  • output: The inspected AI response text, which may be original, masked, or replaced with a block message.
  • prismaAIRSAction: The action taken by Prisma AIRS, typically "block" or other status.
  • response_detected: Details about detected issues in the response.
  • masked_response: Masked version of the response if sensitive data was found.
  • original_response: The original unmodified AI response.
  • response_detection_details: Additional metadata about the detection process.

If a block action is triggered, the output contains the custom block message instead of the original response.

No binary data output is produced by this node.

Dependencies

  • Requires an API key credential for Prisma AIRS AI Security service.
  • Makes HTTP POST requests to the Prisma AIRS API endpoint at https://service.api.aisecurity.paloaltonetworks.com/v1/scan/sync/request.
  • The node expects proper configuration of the Prisma AIRS AI profiles for input or output scanning.
  • Optional metadata like session ID, app user, user IP, app name, and AI model name can be supplied to enrich the request context.

Troubleshooting

  • API Errors: If the node outputs errors like Error calling Prisma AIRS API: ..., verify that the API key credential is valid and has network access to the Prisma AIRS service.
  • Unknown issue output: If the node returns "Unknown issue." in output, it may indicate unexpected or malformed input data; check that required properties are correctly set.
  • Blocking too aggressively: If legitimate prompts or responses are blocked, review and adjust the Prisma AIRS AI profile settings to reduce false positives.
  • Missing required parameters: Ensure that the "Response" and "Prisma AIRS AI Profile for Output" properties are provided when using the Response Inspection operation.
  • Session ID and metadata: While optional, providing consistent session IDs and metadata improves tracking and auditing of inspections.

Links and References

Discussion