Compliance scan

This guide demonstrates how to use the Neural Trust scanning capabilities to analyze your AI system for potential vulnerabilities.

Basic Setup

First, import and initialize the Neural Trust API client:

from neuraltrust import NeuralTrustApi

client = NeuralTrustApi()

Running Scans

Default Scan

The simplest way to run a scan is using the default configuration:

response = client.scanner.scan()

This will:

• Run checks across all available security categories
• Use 5 objectives per category
• Use default attack configurations

Customized Scan

For more control, you can customize your scan parameters:

response = client.scanner.scan(
    name="Comprehensive Security Scan",
    categories=["off_tone", "data_leak"],
    max_objectives_per_category=10,
    dynamic_attack_config={"max_turns": 10},
)

Parameters Explained:

• name: A descriptive name for your scan
• categories: List of specific categories to scan for (e.g., "off_tone", "data_leak")
• max_objectives_per_category: Maximum number of test objectives per category
• dynamic_attack_config: Additional configuration for attack simulations
  • max_turns: Maximum conversation turns for attack attempts

Complete Script

Here's the complete script for reference:

from neuraltrust import NeuralTrustApi

# Initialize the client
client = NeuralTrustApi()

# Default scan, using all categories, 5 objectives per category
response = client.scanner.scan()

# Example using a subset of scan categories, max 10 turns for conversation attack
response = client.scanner.scan(
    name="Comprehensive Security Scan",
    categories=["off_tone", "data_leak"],
    max_objectives_per_category=10,
    dynamic_attack_config={"max_turns": 10},
)