Beyond Vector Hiding: Breaking and Mitigating Shared-Direction Weight Obfuscation in TEE-Offloaded Large Language Models
This work identifies shared rank-one reuse as the root cause of the leakage exploited by the authors' attacks, and proposes LatticeLeak, which exploits the resulting hidden lattice.
Meng-Hui Zhang, Ao-Ying Zheng, Guo-Xiao Liu et al.
· 0 citations