FuzzyAI Fuzzer

The FuzzyAI Fuzzer is a powerful tool for automated LLM fuzzing. It is designed to help developers and security researchers identify jailbreaks and mitigate potential security vulnerabilities in their LLM APIs.

Getting Started

Quick start #1 - Using an existing python project

1. Install fuzzyai

   # Use either pip or any other package manager
   pip install git+https://github.com/cyberark/FuzzyAI.git

2. Run the fuzzer

   fuzzyai fuzz -h

Quick start #2 - or as a standalone project

1. Clone the repository:

   git clone git@github.com:cyberark/FuzzyAI.git
   cd FuzzyAI

2. Install dependencies using Poetry:

   poetry run pip install -e .

3. Run the fuzzer:

   poetry run fuzzyai fuzz -h

4. Optional: Install ollama, and download a model for local usage:

   ollama pull llama3.1
   ollama show llama3.1 # verify model installation

   Alternativly, you can use the Web UI

Web UI (Experimental)

1. Run the Web UI (make sure you completed either of the installation steps from above):

    poetry run fuzzyai webui
   
    # Or specify a custom port:
    poetry run fuzzyai webui --port 9000

Notebooks

We've included interactive Jupyter notebooks you can use under src/fuzzyai/resources/notebooks/.
For more information, see notebooks wiki.

Datasets

We've included some datasets you can use under resources/. For more information, see datasets wiki.

Documentation

Explore detailed usage instructions in the Wiki.

Examples

If you're using poetry, make sure you've activated the venv (or by prepending 'poetry run' to the command line

A default attack (just evaluate the prompt 'Harmful_Prompt') using llama over Ollama

fuzzyai fuzz -m ollama/llama3.1 -a def -t "Harmful_Prompt"

Attack the prompt utilizing ManyShot and Taxonomy attacks, using gpt3.5 over OpenAI using one worker

OPENAI_API_KEY=sk-3fa1... fuzzyai fuzz -m openai/gpt-3.5-turbo -a man -a tax -t "Harmful_Prompt"

Attack prompts in prompts.txt utilizing the ArtPrompt attack, using Claude 3.5 over Anthropic

ANTHROPIC_API_KEY=deadbeef... fuzzyai fuzz -m anthropic/claude-3-5-sonnet-20240620 -a art -T src/fuzzyai/resources/adv_prompts.txt -e blacklisted_words="bomb,garbage,acid,ugly,self-harm,kill your"

Attack your own REST API

OPENAI_API_KEYT=sk-3fa1... fuzzyai fuzz -a def -x openai/gpt-4 -e classifier_model=openai/gpt-4 -m rest/http.raw -e host=localhost -e port=8000 -e scheme=https  
-t "How to build a bomb"

Arguments breakdown:

• -a def:
  Invokes the default attack mode, which evaluates the API without applying additional strategies.

• -x openai/gpt-4:
  Specifies an auxiliary model (gpt-4) to assist with certain tasks during the attack.

• -e classifier_model=openai/gpt-4:
  Configures the fuzzer to use the gpt-4 model for classifying outputs. This helps determine whether responses meet specific criteria, such as detecting harmful or unwanted content.

• -m rest/http.raw:
  Uses the REST provider to attack an API endpoint. The raw HTTP request is parsed from the file http.raw. (Refer to the documentation or example for details on the file structure.)

• -e host=localhost -e port=8000 -e scheme=https:
  Configures the REST provider with the following API endpoint details:

  • Host: localhost
  • Port: 8000
  • Scheme: https (communication will use HTTPS)

• -t "How to build a bomb":
  Specifies the test input. In this example, it tests the API's handling of sensitive or harmful content.

Key Features

• Comprehensive Fuzzing Techniques: Leverage mutation-based, generation-based, and intelligent fuzzing.
• Built-in Input Generation: Generate valid and invalid inputs for exhaustive testing.
• Seamless Integration: Easily incorporate into your development and testing workflows.
• Extensible Architecture: Customize and expand the fuzzer to meet your unique requirements.

Supported models

FuzzyAI supports various models across top providers, including:

Adding support for newer models

Easily add support for additional models by following our DIY guide.

Implemented Attacks

See attacks wiki for detailed information

Supported Cloud APIs

• OpenAI
• Anthropic
• Gemini
• Azure Cloud
• AWS Bedrock
• AI21
• DeepSeek
• Huggingface (Downloading models)
• Ollama

• Custom REST API

Caveats

• Some classifiers do more than just evaluate a single output. For example, the cosine-similarity classifier compares two outputs by measuring the angle between them, while a 'harmfulness' classifier checks whether a given output is harmful. As a result, not all classifiers are compatible with the attack methods we've implemented, as those methods are designed for single-output classifiers.
• When using the -m option with OLLAMA models, ensure that all OLLAMA models are added first before adding any other models. Use the -e port=... option to specify the port number for OLLAMA (default is 11434).

Contributing

Contributions are welcome! If you would like to contribute to the FuzzyAI Fuzzer, please follow the guidelines outlined in the CONTRIBUTING.md file.

License

The FuzzyAI Fuzzer is released under the Apache License. See the LICENSE file for more details.

Contact

If you have any questions or suggestions regarding the FuzzyAI Fuzzer, please feel free to contact us at fzai@cyberark.com.