Skip to main content
AI Attack Simulation is a comprehensive platform for testing and strengthening your AI system’s security defenses. AI Attack Simulation provides automated security testing for your AI systems through two complementary approaches:
  • System Prompt Evaluation: Analyze and enhance your system prompts with AI-provided suggestions.
    • Establishes a security baseline
    • Identifies vulnerabilities in system prompts
    • Provides an enhanced prompt for testing
    • Educational - learn security best practices
  • Red Team Evaluation: Run adversarial attacks to identify vulnerabilities in your AI systems.
    • Validates that security controls actually work
    • Identifies remaining vulnerabilities
    • Provides quantifiable security metrics
    • Demonstrates improvement over baseline
Together, these evaluation types help you build robust, secure AI applications. You can run attack simulations and system prompt evaluations in the HiddenLayer Console against a variety of foundation model providers. See System Prompt and Red Teaming for Console workflows. A large part of the value of these tools also lies in configuring attacks against custom applications, not just foundational models. Attack simulations and system prompt evaluations can be performed via API against any application that is accessible through the internet. This section shows you how to configure those simulations using the HiddenLayer SDK.
Accessing ApplicationsAI Attack Simulation is only available as a SaaS offering from HiddenLayer. In order to test custom applications, those applications must have a publicly exposed interface (endpoint, widget, etc.) such that they can be reached by our “attacker”, and must be able to return a response back to the testing endpoint.
Running simulations against your custom applications programmatically allows you to:
  • Establish a security baseline
  • Identify vulnerabilities in application stacks, not just in their component parts
  • Track and compare quantifiable security metrics
  • Integrate evaluations and simulations as part of the development and release pipeline
This approach helps secure AI applications throughout their development lifecycle; allows teams to create a baseline for acceptable security standards; and ensures that application security remains high throughout the application’s life.