#adversarial prompts

A new method allows rewriting blocked prompts in text-to-video systems to bypass safety filters without changing meaning, revealing weaknesses in current AI content safeguards.