The rapid progress in open-source large language models (LLMs) is
significantly advancing AI development. Extensive efforts have been made before
model release to align their behavior with human values, with the primary goal
of ensuring their helpfulness and harmlessness. However, even carefully aligned
models can be manipulated maliciously, leading to unintended behaviors, known
as "jailbreaks". These jailbreaks are typically triggered by specific text
inputs, often referred to as adversarial prompts. In this work, we propose the
generation exploitation attack, an extremely simple approach that disrupts
model alignment by only manipulating variations of decoding methods. By
exploiting different generation strategies, including varying decoding
hyper-parameters and sampling methods, we increase the misalignment rate from
0% to more than 95% across 11 language models including LLaMA2, Vicuna, Falcon,
and MPT families, outperforming state-of-the-art attacks with $30\times$ lower
computational cost. Finally, we propose an effective alignment method that
explores diverse generation strategies, which can reasonably reduce the
misalignment rate under our attack. Altogether, our study underscores a major
failure in current safety evaluation and alignment procedures for open-source
LLMs, strongly advocating for more comprehensive red teaming and better
alignment before releasing such models. Our code is available at
this https URL

通过改变文本生成策略，我们提出一种新的攻击方法（生成利用攻击），成功地使 11 种语言模型的对齐率从 0% 提高到超过 95%；我们还提出了一种有效的对齐方法，可在攻击下合理降低对齐失误率。这项研究强烈呼吁在发布开源大型语言模型之前进行更全面的红队测试和更好的对齐方式。