On the Safety of Open-Sourced Large Language Models: Does Alignment Really Prevent Them From Being Misused?
This reveals a critical safety vulnerability in open-sourced LLMs, highlighting the need for more advanced mitigation strategies to prevent misuse.
The paper tackles the problem of whether alignment techniques like SFT or RLHF can prevent open-sourced large language models from generating undesired content, showing that they can be easily misused to produce harmful, biased, or private information without heavy computations or careful prompts.
Large Language Models (LLMs) have achieved unprecedented performance in Natural Language Generation (NLG) tasks. However, many existing studies have shown that they could be misused to generate undesired content. In response, before releasing LLMs for public access, model developers usually align those language models through Supervised Fine-Tuning (SFT) or Reinforcement Learning with Human Feedback (RLHF). Consequently, those aligned large language models refuse to generate undesired content when facing potentially harmful/unethical requests. A natural question is "could alignment really prevent those open-sourced large language models from being misused to generate undesired content?''. In this work, we provide a negative answer to this question. In particular, we show those open-sourced, aligned large language models could be easily misguided to generate undesired content without heavy computations or careful prompt designs. Our key idea is to directly manipulate the generation process of open-sourced LLMs to misguide it to generate undesired content including harmful or biased information and even private data. We evaluate our method on 4 open-sourced LLMs accessible publicly and our finding highlights the need for more advanced mitigation strategies for open-sourced LLMs.