TYPO: Instruction-Dense Visual Jailbreaks against Commercial Closed-Source Image-Generation Models

Recent commercial image-generation models can generate high-quality images with readable text e.g., posters, infographics, and manuals, attracting considerable attention. Yet we first show that this same capability also introduces a previously unreported safety vulnerability: these systems may...

MAI-2024-0005

The Chain-of-Jailbreak CoJ attack is a sophisticated method designed to circumvent safety protocols in image generation models. This attack operates by fragmenting a malicious query into a series of innocuous sub-queries. Each sub-query prompts the model to incrementally edit the image, ultimatel...