Sandwich attack: Multi-language Mixture Adaptive Attack on LLMs

arXiv:2404.07242v1 Announce Type: new
Abstract: Large Language Models (LLMs) are increasingly being developed and applied, but their widespread use faces challenges. These include aligning LLMs' responses with human values to prevent harmful outputs, which is addressed through safety training methods. Even so, bad actors and malicious users have succeeded in attempts to manipulate the LLMs to generate misaligned responses for harmful questions such as methods to create a bomb in school labs, recipes for harmful drugs, and ways to evade …

arxiv attack bad bad actors challenges cs.ai cs.cl cs.cr human human values language language models large llms malicious prevent safety sandwich training