mlsanityData to 14 Oct 2026

Mask2Shield: Strengthening LLM Safety against Neuron-Pruning Attacks

arXiv 2607.23015 ↗First posted 25 Jul 2026cs.CRcited by 0

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

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 recover a safe refusal through the remaining computation, while a frozen, unmasked teacher supplies complete benign answers to limit capability drift. Across ten model configurations, M2S reduces successful recomputed pruning attacks from 80--279 to 1--44 out of 313 prompts while generally preserving four capability benchmarks. We also evaluate M2S with TwinBreak, which uses a different neuron-selection rule and iterative pruning procedure. Together, these results show that M2S makes targeted pruning less effective by reducing reliance on a small, removable safety-neuron set.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 26 of its 30 references that are in the map. First posted on arXiv 25 Jul 2026.

Builds on

26 of 30 references.

Closest ancestors · 6

Same direction, newest first

  1. 2026SafeNeuron: Neuron-Level Safety Alignment for Large Language Models12 Feb
  2. 2025NeuroStrike: Neuron-Level Attacks on Aligned LLMs15 Sep
  3. 2025NeuronTune: Fine-Grained Neuron Modulation for Balanced Safety-Utility Alignment in LLMs13 Aug
  4. 2025NeuRel-Attack: Neuron Relearning for Safety Disalignment in Large Language Models29 Apr
  5. 2024Refusal in Language Models Is Mediated by a Single Direction17 Jun
  6. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!5 Oct

Other references · 20

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  5. 2024Qwen2.5 Technical Reportcited by 4,942
  6. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,886
  7. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  8. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 3,050
  9. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  10. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  11. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,420
  12. 2024Gemma: Open Models Based on Gemini Research and Technologycited by 916
8 more
  1. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 785
  2. 2024Phi-4 Technical Reportcited by 711
  3. 2024JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Modelscited by 536
  4. 2024A StrongREJECT for Empty Jailbreakscited by 385
  5. 2024ShortGPT: Layers in Large Language Models are More Redundant Than You Expectcited by 383
  6. 2024BlockPruner: Fine-grained Pruning for Large Language Modelscited by 36
  7. 2025TwinBreak: Jailbreaking LLM Security Alignments based on Twin Promptscited by 19
  8. 2025Beyond Surface Alignment: Rebuilding LLMs Safety Mechanism via Probabilistically Ablating Refusal Directioncited by 8

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026NeuronGuard: Robust LLM Safety Alignment via Ablation-Aware Safety Signal Redistribution
  2. 2026NeST: Neuron Selective Tuning for LLM Safety
  3. 2024Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications
  4. 2026A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models
  5. 2026Light Alignment Improves LLM Safety via Model Self-Reflection with a Single Neuron
  6. 2026Faithful Dual-constrained Erasure for Robust LLM Safety Alignment
  7. 2024NLSR: Neuron-Level Safety Realignment of Large Language Models Against Harmful Fine-Tuning
  8. 2025Improving LLM Safety Alignment with Dual-Objective Optimization