CNT: Safety-oriented Function Reuse across LLMs via Cross-Model Neuron Transfer

2026-03-18

AI Summary

This work addresses the challenge of flexibly and efficiently transferring safety capabilities from open-source large language models to meet dynamic safety requirements without retraining. The authors propose Cross-model Neuron Transfer (CNT), a novel approach that, for the first time, enables modular reuse of safety functionalities at the neuron level across different models. CNT selectively transfers a minimal subset of neurons from a donor model to a target model and integrates them via posterior functional adaptation and modular network editing, thereby supporting plug-and-play deployment and removal of safety features. Experiments across seven prominent large language models demonstrate that CNT achieves high transfer success rates in tasks including safety de-alignment, alignment enhancement, and bias mitigation, with performance degradation of less than 1%, significantly outperforming five baseline methods.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

The widespread deployment of large language models (LLMs) calls for post-hoc methods that can flexibly adapt models to evolving safety requirements. Meanwhile, the rapidly expanding open-source LLM ecosystem has produced a diverse collection of models that already exhibit various safety-related functionalities. This motivates a shift from constructing safety functionality from scratch to reusing existing functionality from external models, thereby avoiding costly data collection and training procedures. In this paper, we present Cross-Model Neuron Transfer (CNT), a post-hoc method that reuses safety-oriented functionality by transferring a minimal subset of neurons from an open-source donor LLM to a target LLM. By operating at the neuron level, CNT enables modular function-level adaptation, supporting both function addition andfunction deletion. We evaluate CNT on seven popular LLMs across three representative applications: safety disalignment, alignment enhancement, and bias removal. Experimental results show that CNT achieves targeted safety-oriented functionality transfer with minimal performance degradation (less than 1% for most models), consistently outperforming five baselines, demonstrating its generality and practical effectiveness.

Problem

• safety-oriented function reuse
• large language models
• cross-model transfer
• post-hoc adaptation
• neuron-level transfer

Innovation

• Cross-Model Neuron Transfer
• Safety Alignment
• Function Reuse
• Modular Adaptation
• Post-hoc Safety Intervention