Typographic tricks let harmful text bypass AI content filters

Original: What the Eyes See, the LLMs Miss: Exploiting Human Perception for Adversarial Text Attacks

Source: arxiv.org ↗