FuzzyAI
StaleDescription
An automated LLM fuzzing tool by CyberArk that helps developers and security researchers identify and mitigate jailbreak vulnerabilities in LLM APIs with multiple attack vectors.
Key Features
- Comprehensive fuzzing techniques — mutation-based, generation-based, and intelligent fuzzing for LLM APIs
- Multiple attack strategies — includes default, ManyShot, Taxonomy, ArtPrompt, and other jailbreak attack vectors
- Multi-provider support — works with OpenAI, Anthropic, Gemini, Azure, Bedrock, AI21, DeepSeek, and Ollama
- REST API fuzzing — test your own custom REST API endpoints with configurable host, port, and scheme
- Built-in web UI — experimental graphical interface for interactive fuzzing sessions
- Extensible architecture — easily add new models, attack types, and custom classifiers
Use Cases
💡 LLM security auditing — identifying jailbreak vulnerabilities before deploying AI models to production
💡 Red team testing — automated adversarial testing of AI system defenses against prompt injection attacks
💡 Compliance validation — verifying that AI APIs properly reject harmful or policy-violating inputs
💡 Model evaluation — benchmarking the robustness of different LLM providers against known attack patterns
Strengths & Limitations
✅ Strengths
- • High community interest (1.6k stars)
- • Permissive open-source license (Apache-2.0)
- • Responsive to issues, low backlog
⚠️ Limitations
- • No updates in over 6 months
Categories
Quick Start
pip install git+https://github.com/cyberark/FuzzyAI.git → fuzzyai fuzz -h for help → fuzzyai fuzz -m openai/gpt-4o -a def -t 'your test prompt' → Add -a man -a tax for multiple attack vectors → Use fuzzyai webui for graphical interface