CNT: Safety-oriented Function Reuse across LLMs via Cross-Model Neuron Transfer
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915874276900864 |
|---|---|
| author | Zhao, Yue Gong, Yujia Liang, Ruigang Zhu, Shenchen Chen, Kai Yuan, Xuejing Zhang, Wangjun |
| author_facet | Zhao, Yue Gong, Yujia Liang, Ruigang Zhu, Shenchen Chen, Kai Yuan, Xuejing Zhang, Wangjun |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_18449 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | CNT: Safety-oriented Function Reuse across LLMs via Cross-Model Neuron Transfer Zhao, Yue Gong, Yujia Liang, Ruigang Zhu, Shenchen Chen, Kai Yuan, Xuejing Zhang, Wangjun Cryptography and Security Software Engineering 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. |
| title | CNT: Safety-oriented Function Reuse across LLMs via Cross-Model Neuron Transfer |
| topic | Cryptography and Security Software Engineering |
| url | https://arxiv.org/abs/2603.18449 |