ArtPrompt : An ASCII art based vulnerability in LLMs

With the growing trend of Generative AI, particularly LLMs like GPT, Claude, Gemini, and Llama2, it also brings the new challenges, especially in the area of security. Security attacks on Large Language Models (LLMs) exploit their vulnerabilities to induce harmful behaviors. A recent study has shed light on new vulnerability known as “ArtPrompt” that uses ASCII art to bypass LLM safety measures.