AI Agent Security for Low-Code/No-Code Platforms
Focus
Focus
Prisma AIRS

AI Agent Security for Low-Code/No-Code Platforms

Table of Contents

AI Agent Security for Low-Code/No-Code Platforms

Secure your AI agents running on low-code or no-code platforms
Where Can I Use This?What Do I Need?
  • AI Agent Security
Prisma AIRS detects and prevents malicious exploitation of AI agents deployed on no-code/low-code platforms, protecting against unauthorized actions and system manipulation.
The AI Agent security provides model-based and pattern-based detection services.
AI agents could be deployed on no-code/low-code platforms like AWS Agent Builder, GCP Agent Builder, Microsoft Copilot Studio, and Azure AI Agent Builder.
The feature secures your AI Agents using a prompt injection detection service against threat vectors such as:
  • Function schema extraction: Leaks when asking questions about available tools/schema.
  • Direct function invocation: Involves leakage of tool names; the attacker asks the agent to invoke the tool directly.
  • Memory manipulation: includes directly modifying long-term memory when incorporating predefined schema tags.
AI Agent Security Components:
  • AI Agent framework: The framework in which your cloud workloads run.
  • AI Agent protection: Update API security profile to activate agent-specific threat detection. When enabled, the system will inspect API calls for signs of function schema extraction, direct function invocation, and memory manipulation attempts.
To secure your AI agents,
  1. Configure a Prisma AIRS API security profile by defining applications and configuring an AI agent framework.
  2. Enable AI Agent Protection in your API security profile.
    If you enable AI Agent Protection without configuring an AI Agent framework in your application definition, then the AI Agent detection service only enables model-based protections and not pattern-based.
  3. Review the API scan logs for any agent-related threats identified by fields such as Agent framework, Agent final verdict, prompt Agent verdict, and prompt Agent action.
  4. See the Prisma AIRS Scan APIs to detect agent-specific threats in the scan prompt and response.