Guy Amit (Ben-Gurion University), Moshe Levy (Ben-Gurion University), Yisroel Mirsky (Ben-Gurion University)

Deep neural networks are normally executed in the forward direction. However, in this work, we identify a vulnerability that enables models to be trained in both directions and on different tasks. Adversaries can exploit this capability to hide rogue models within seemingly legitimate models. In addition, in this work we show that neural networks can be taught to systematically memorize and retrieve specific samples from datasets. Together, these findings expose a novel method in which adversaries can exfiltrate datasets from protected learning environments under the guise of legitimate models.

We focus on the data exfiltration attack and show that modern architectures can be used to secretly exfiltrate tens of thousands of samples with high fidelity, high enough to compromise data privacy and even train new models. Moreover, to mitigate this threat we propose a novel approach for detecting infected models.

View More Papers

Analysis of the Effect of the Difference between Japanese...

Rei Yamagishi, Shinya Sasa, and Shota Fujii (Hitachi, Ltd.)

Read More

Space-Domain AI Applications need Rigorous Security Risk Analysis

Alexandra Weber (Telespazio Germany GmbH), Peter Franke (Telespazio Germany GmbH)

Read More

BreakSPF: How Shared Infrastructures Magnify SPF Vulnerabilities Across the...

Chuhan Wang (Tsinghua University), Yasuhiro Kuranaga (Tsinghua University), Yihang Wang (Tsinghua University), Mingming Zhang (Zhongguancun Laboratory), Linkai Zheng (Tsinghua University), Xiang Li (Tsinghua University), Jianjun Chen (Tsinghua University; Zhongguancun Laboratory), Haixin Duan (Tsinghua University; Quan Cheng Lab; Zhongguancun Laboratory), Yanzhong Lin (Coremail Technology Co. Ltd), Qingfeng Pan (Coremail Technology Co. Ltd)

Read More

Work-in-Progress: A Large-Scale Long-term Analysis of Online Fraud across...

Yi Han, Shujiang Wu, Mengmeng Li, Zixi Wang, and Pengfei Sun (F5)

Read More