Tag
This paper investigates the safety of large language models (LLMs) beyond text inputs by examining emoji-augmented prompts, revealing gaps in current safety evaluations and model-dependent vulnerabilities.