Wendi Feng

Hao Li, Jingkun An, Zijun Song et al.

Aligning Large Language Models (LLMs) with human values often degrades their general capabilities, termed the alignment tax. Existing methods mitigate this by balancing dual objectives, which heavily rely on massive general-purpose data or auxiliary reward models. In this paper, we argue that, because safety features are inherently sparse within the output distribution, alignment requires localized modifications rather than global trade-offs. To this end, we propose SafeSteer, which performs on-policy distillation confined to safety tokens. First, we construct a safety teacher via activation steering. Based on this teacher, we develop a safety token selection algorithm. Consequently, SafeSteer restricts the reverse KL penalty to these tokens during training to preserve general capabilities. Experimental results across diverse models show that our SafeSteer achieves a superior trade-off between safety and general capability compared with existing methods, attaining strong safety performance on seven safety benchmarks with only minimal degradation on five general capability benchmarks. Notably, SafeSteer requires only 100 harmful samples without using any general-purpose data, less than 1% of what previous baselines used, considerably reducing alignment cost. More details are on our project page at https://anjingkun.github.io/SafeSteer.

Wendi Feng, Chuanchang Liu, Zehua Guo et al.

Sensitive data protection is essential for mobile users. Plausibly Deniable Encryption (PDE) systems provide an effective manner to protect sensitive data by hiding them on the device. However, existing PDE systems can lose data due to overriding the hidden volume, waste physical storage because of the reserved area used for avoiding data loss, and require device reboot when using the hidden volume. This paper presents MobiGyges, a hidden volume-based mobile PDE system, to fill the gap. MobiGyges addresses the problem of data loss by restricting each storage block used only by one volume, and it improves storage utilization by eliminating the reserved area. MobiGyges can also avoid device reboot by mounting the hidden volume dynamically on-demand with the Dynamic Mounting service. Moreover, we identify two novel PDE oriented attacks, the capacity comparison attack and the fill-to-full attack. MobiGyges can defend them by jointly leveraging the Shrunk U-disk method and multi-level deniability. We implement the MobiGyges proof-of-concept system on a real mobile phone Google Nexus 6P with LineageOS 13. Experimental results show that MobiGyges prevents data loss, avoids device reboot, improves storage utilization by over 30% with acceptable performance overhead compared with current works.