Shang Shang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Shang Shang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Liya Su
Institute of Information Engineering Chinese Academy of Sciences
2 shared publications - Xiaodan Zhang
Chinese Academy of Sciences
2 shared publications - Yepeng Yao
Institute of Information Engineering , Chinese Academy of Sciences, Beijing, 100085
2 shared publications - Zhengwei Jiang
Institute of Information Engineering, Chinese Academy of Sciences
2 shared publications - Zhongjiang Yao
Institute of Information Engineering Chinese Academy of Sciences
2 shared publications - Zijing Fan
Institute of Information Engineering Chinese Academy of Sciences
2 shared publications
Latest Publications
- 5 September 2024
- Published by Springer Nature
- p. 146-165
IntentObfuscator: A Jailbreaking Method via Confusing LLM with Prompts
- 5 September 2024
- Published by Springer Nature
- p. 146-165
IntentObfuscator: A Jailbreaking Method via Confusing LLM with Prompts
- Open Access
Can LLMs Deeply Detect Complex Malicious Queries? A Framework for Jailbreaking via Obfuscating Intent
To demonstrate and address the underlying maliciousness, we propose a theoretical hypothesis and analytical approach, and introduce a new black-box jailbreak attack methodology named IntentObfuscator, exploiting this identified flaw by obfuscating the true intentions behind user prompts.This approach compels LLMs to inadvertently generate restricted content, bypassing their built-in content security measures. We detail two implementations under this framework: "Obscure Intention" and "Create Ambiguity", which manipulate query complexity and ambiguity to evade malicious intent detection effectively. We empirically validate the effectiveness of the IntentObfuscator method across several models, including ChatGPT-3.5, ChatGPT-4, Qwen and Baichuan, achieving an average jailbreak success rate of 69.21\%. Notably, our tests on ChatGPT-3.5, which claims 100 million weekly active users, achieved a remarkable success rate of 83.65\%. We also extend our validation to diverse types of sensitive content like graphic violence, racism, sexism, political sensitivity, cybersecurity threats, and criminal skills, further proving the substantial impact of our findings on enhancing 'Red Team' strategies against LLM content security frameworks.
Open AccessCan LLMs Deeply Detect Complex Malicious Queries? A Framework for Jailbreaking via Obfuscating Intent
To demonstrate and address the underlying maliciousness, we propose a theoretical hypothesis and analytical approach, and introduce a new black-box jailbreak attack methodology named IntentObfuscator, exploiting this identified flaw by obfuscating the true intentions behind user prompts.This approach compels LLMs to inadvertently generate restricted content, bypassing their built-in content security measures. We detail two implementations under this framework: "Obscure Intention" and "Create Ambiguity", which manipulate query complexity and ambiguity to evade malicious intent detection effectively. We empirically validate the effectiveness of the IntentObfuscator method across several models, including ChatGPT-3.5, ChatGPT-4, Qwen and Baichuan, achieving an average jailbreak success rate of 69.21\%. Notably, our tests on ChatGPT-3.5, which claims 100 million weekly active users, achieved a remarkable success rate of 83.65\%. We also extend our validation to diverse types of sensitive content like graphic violence, racism, sexism, political sensitivity, cybersecurity threats, and criminal skills, further proving the substantial impact of our findings on enhancing 'Red Team' strategies against LLM content security frameworks.