Jailbreaking Text-to-Video Systems with Rewritten Prompts

Researchers have developed a new method to bypass safety filters in text-to-video systems without changing the original meaning of prompts. The approach was tested on platforms like Pika, Luma, Kling, and Open-Sora, consistently outperforming previous techniques. By rewriting blocked prompts to preserve semantics while avoiding detection, the method achieved high success rates in generating restricted content. This reveals significant vulnerabilities in current safety mechanisms for text-to-video models, highlighting the urgent need for more robust defenses against potential misuse.