Configure Validation

Validation Process

Validation simulates adversarial behavior by generating and executing thousands of attack prompts across multiple threat vectors and malicious intent categories. The system under test is then analyzed based on its responses to identify vulnerabilities, including security weaknesses, privacy risks, and safety concerns. A well-secured AI system will mitigate these attacks, typically by rejecting unsafe requests.

For example, responding with :

I'm sorry, I cannot assist with that task.

Set up and Run Validation

To set up Validation and assess the security posture of your AI models and applications, follow these steps:

1. Connect to Your LLM API Provider

   • Follow the instructions in LLM API Provider integrations to establish a connection with the provider hosting your AI models.

2. Integrate AI Defense with AWS Bedrock (or Your Chosen Provider)

   • Navigate to the AI Defense Administration tab.

   • Locate the AWS Bedrock card and click Connect.

   • Enter the necessary API key details to complete the integration.

   • For additional guidance, refer to the AI Defense Administration documentation.

3. Ensure Multicloud Defense is Connected

   • Check the Multicloud Defense card on the Administration tab.

   • If a Disconnect button is visible, the connection is active.

   • If not, follow the Set up Asset Discovery instructions to enable it.

4. Add Assets for Validation

   • Follow the steps in Find Asset and Add an Asset to register the models and applications you want to test.

After Running a Validation Test

Once Validation is configured and executed, you can:

• Analyze test results to identify vulnerabilities in your AI system.

• Adjust model configurations, update policies, or fine-tune prompts based on findings.

• Use Runtime or Runtime Gateway to enforce stricter security measures.

• Run another validation test to ensure fixes are effective and threats are mitigated.

This structured approach helps continuously strengthen AI security, ensuring compliance with industry standards like OWASP Top 10 for LLMs and MITRE ATLAS.