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

What Makes Phishing Simulation Campaigns (Un)Acceptable? A Vignette Experiment

Jasmin Schwab (German Aerospace Center (DLR)), Alexander Nussbaum (University of the Bundeswehr Munich), Anastasia Sergeeva (University of Luxembourg), Florian Alt (University of the Bundeswehr Munich and Ludwig Maximilian University of Munich), and Verena Distler (Aalto University)

Read More

Truman: Constructing Device Behavior Models from OS Drivers to...

Zheyu Ma (Institute for Network Sciences and Cyberspace (INSC), Tsinghua University; EPFL; JCSS, Tsinghua University (INSC) - Science City (Guangzhou) Digital Technology Group Co., Ltd.), Qiang Liu (EPFL), Zheming Li (Institute for Network Sciences and Cyberspace (INSC), Tsinghua University; JCSS, Tsinghua University (INSC) - Science City (Guangzhou) Digital Technology Group Co., Ltd.), Tingting Yin (Zhongguancun…

Read More

AegisSat: A Satellite Cybersecurity Testbed

Roee Idan, Roy Peled, Aviel Ben Siman Tov, Eli Markus, Boris Zadov, Ofir Chodeda, Yohai Fadida (Ben Gurion University of the Negev), Oliver Holschke, Jan Plachy (T-Labs (Research & Innovation)), Yuval Elovici, Asaf Shabtai (Ben Gurion University of the Negev)

Read More

ASGARD: Protecting On-Device Deep Neural Networks with Virtualization-Based Trusted...

Myungsuk Moon (Yonsei University), Minhee Kim (Yonsei University), Joonkyo Jung (Yonsei University), Dokyung Song (Yonsei University)

Read More