Grey Box Testing

Grey-box AI Red Teaming involves testing with partial knowledge of the system, such as knowing the model type (e.g., GPT-4), having access to some documentation, or understanding the general system architecture but not having full model weights or source code. This allows for more targeted testing than black-box while still simulating realistic external attacker scenarios where some information might be gleaned.

Learn more from the following resources:

• @article@AI Transparency: Connecting AI Red Teaming and Compliance
• @article@Black-Box, Gray Box, and White-Box Penetration Testing
• @article@Understanding Black Box, White Box, and Grey Box Testing