Agentic Security
An open-source LLM vulnerability scanner tests text, image and audio attacks against model APIs and supports automated checks in GitHub Actions.
Agentic Security is an LLM vulnerability scanner for developers, researchers and security teams testing AI models and agent workflows. It probes how systems respond to hostile inputs and attempts to bypass their safety controls.
- Multimodal attacks: Test model behavior with text, images and audio, extending security checks beyond text prompts.
- Adaptive jailbreaks: Combine multi-step attacks, randomized inputs and reinforcement learning based probes to look for weaknesses in model defenses.
- Custom test coverage: Bring your own prompt datasets, generate variations through mutations, and tailor testing with custom rule sets or agent based attacks.
- API stress testing: Send high volumes of attack inputs to LLM APIs to examine unexpected responses under load.
- Development checks: Run automated scans in CI/CD through GitHub Actions, with a UI also available for working with the scanner.
The scanner is written in Python and is open source under Apache 2.0. Tests send attack inputs to the model endpoint you choose; requests to external model services leave your machine. It identifies potential weaknesses but doesn't guarantee protection against every threat.
Logotyper
