Mask2Shield: Strengthening LLM Safety against Neuron-Pruning Attacks

본문 미리보기

arXiv:2607.23015v1 Announce Type: new Abstract: Large language models (LLMs) are safety-aligned before deployment to reduce harmful content generation. Yet neuron-level pruning attacks show that refusal can depend on a small set of removable units: disabling them can remove safety behavior while leaving much of the model usable. To address this problem, we introduce Mask2Shield (M2S), a masked-forward alignment method that trains a model under this functional pruning. The masked student must re