Yichen Gong (Tsinghua University), Delong Ran (Tsinghua University), Xinlei He (Hong Kong University of Science and Technology (Guangzhou)), Tianshuo Cong (Tsinghua University), Anyu Wang (Tsinghua University), Xiaoyun Wang (Tsinghua University)

The safety alignment of Large Language Models (LLMs) is crucial to prevent unsafe content that violates human values.
To ensure this, it is essential to evaluate the robustness of their alignment against diverse malicious attacks.
However, the lack of a large-scale, unified measurement framework hinders a comprehensive understanding of potential vulnerabilities.
To fill this gap, this paper presents the first comprehensive evaluation of existing and newly proposed safety misalignment methods for LLMs. Specifically, we investigate four research questions: (1) evaluating the robustness of LLMs with different alignment strategies, (2) identifying the most effective misalignment method, (3) determining key factors that influence misalignment effectiveness, and (4) exploring various defenses.
The safety misalignment attacks in our paper include system-prompt modification, model fine-tuning, and model editing.
Our findings show that Supervised Fine-Tuning is the most potent attack but requires harmful model responses.
In contrast, our novel Self-Supervised Representation Attack (SSRA) achieves significant misalignment without harmful responses.
We also examine defensive mechanisms such as safety data filter, model detoxification, and our proposed Self-Supervised Representation Defense (SSRD), demonstrating that SSRD can effectively re-align the model.
In conclusion, our unified safety alignment evaluation framework empirically highlights the fragility of the safety alignment of LLMs.

View More Papers

The Discriminative Power of Cross-layer RTTs in Fingerprinting Proxy...

Diwen Xue (University of Michigan), Robert Stanley (University of Michigan), Piyush Kumar (University of Michigan), Roya Ensafi (University of Michigan)

Read More

SCAMMAGNIFIER: Piercing the Veil of Fraudulent Shopping Website Campaigns

Marzieh Bitaab (Arizona State University), Alireza Karimi (Arizona State University), Zhuoer Lyu (Arizona State University), Adam Oest (Amazon), Dhruv Kuchhal (Amazon), Muhammad Saad (X Corp.), Gail-Joon Ahn (Arizona State University), Ruoyu Wang (Arizona State University), Tiffany Bao (Arizona State University), Yan Shoshitaishvili (Arizona State University), Adam Doupé (Arizona State University)

Read More

MineShark: Cryptomining Traffic Detection at Scale

Shaoke Xi (Zhejiang University), Tianyi Fu (Zhejiang University), Kai Bu (Zhejiang University), Chunling Yang (Zhejiang University), Zhihua Chang (Zhejiang University), Wenzhi Chen (Zhejiang University), Zhou Ma (Zhejiang University), Chongjie Chen (HANG ZHOU CITY BRAIN CO., LTD), Yongsheng Shen (HANG ZHOU CITY BRAIN CO., LTD), Kui Ren (Zhejiang University)

Read More

Alba: The Dawn of Scalable Bridges for Blockchains

Giulia Scaffino (TU Wien), Lukas Aumayr (TU Wien), Mahsa Bastankhah (Princeton University), Zeta Avarikioti (TU Wien), Matteo Maffei (TU Wien)

Read More