Chengkun Wei (Zhejiang University), Wenlong Meng (Zhejiang University), Zhikun Zhang (CISPA Helmholtz Center for Information Security and Stanford University), Min Chen (CISPA Helmholtz Center for Information Security), Minghu Zhao (Zhejiang University), Wenjing Fang (Ant Group), Lei Wang (Ant Group), Zihui Zhang (Zhejiang University), Wenzhi Chen (Zhejiang University)

*Prompt-tuning* has emerged as an attractive paradigm for deploying large-scale language models due to its strong downstream task performance and efficient multitask serving ability. Despite its wide adoption, we empirically show that prompt-tuning is vulnerable to downstream task-agnostic backdoors, which reside in the pretrained models and can affect arbitrary downstream tasks. The state-of-the-art backdoor detection approaches cannot defend against task-agnostic backdoors since they hardly converge in reversing the backdoor triggers. To address this issue, we propose LMSanitator, a novel approach for detecting and removing task-agnostic backdoors on Transformer models. Instead of directly inverting the triggers, LMSanitator aims to invert the *predefined attack vectors* (pretrained models' output when the input is embedded with triggers) of the task-agnostic backdoors, which achieves much better convergence performance and backdoor detection accuracy. LMSanitator further leverages prompt-tuning’s property of freezing the pretrained model to perform accurate and fast output monitoring and input purging during the inference phase. Extensive experiments on multiple language models and NLP tasks illustrate the effectiveness of LMSanitator. For instance, LMSanitator achieves 92.8% backdoor detection accuracy on 960 models and decreases the attack success rate to less than 1% in most scenarios.

View More Papers

5G-Spector: An O-RAN Compliant Layer-3 Cellular Attack Detection Service

Haohuang Wen (The Ohio State University), Phillip Porras (SRI International), Vinod Yegneswaran (SRI International), Ashish Gehani (SRI International), Zhiqiang Lin (The Ohio State University)

Read More

Towards Precise Reporting of Cryptographic Misuses

Yikang Chen (The Chinese University of Hong Kong), Yibo Liu (Arizona State University), Ka Lok Wu (The Chinese University of Hong Kong), Duc V Le (Visa Research), Sze Yiu Chau (The Chinese University of Hong Kong)

Read More

From Interaction to Independence: zkSNARKs for Transparent and Non-Interactive...

Shahriar Ebrahimi (IDEAS-NCBR), Parisa Hassanizadeh (IDEAS-NCBR)

Read More

Efficient and Timely Revocation of V2X Credentials

Gianluca Scopelliti (Ericsson & KU Leuven), Christoph Baumann (Ericsson), Fritz Alder (KU Leuven), Eddy Truyen (KU Leuven), Jan Tobias Mühlberg (Université libre de Bruxelles & KU Leuven)

Read More