The sample applications here are deliberately insecure to demonstrate Agent security risks. Please exercise caution when deploying in your environment.
We aim to demonstrate security risks in well-known Agentic AI (also known as Agents) frameworks, particularly how Agent misconfigurations (i.e., insecure code, framework-specific misconfigurations) can lead to vulnerabilities identified in the OWASP Top 10 for Gen AI and Large Language Model Applications 2025.
This repository contains examples of insecure code and/or security misconfigurations in common Agent frameworks such as:
More Agent frameworks will be supported in the future.
The framework folder contains subdirectories for each framework with individual examples of vulnerable agents. Each example is accompanied by a description of the vulnerability.
The guidelines for contributing are described in the CONTRIBUTING.md file.
Visit the ASI project landing page: (https://genai.owasp.org/initiatives/#agenticinitiative)