![[2602.16835] NeST: Neuron Selective Tuning for LLM Safety](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16835] NeST: Neuron Selective Tuning for LLM Safety
Summary
The paper introduces NeST, a novel framework for enhancing safety in large language models (LLMs) by selectively tuning a small subset of neurons, significantly reducing unsafe outputs while minimizing computational overhead.
Why It Matters
As LLMs become more prevalent, ensuring their safe deployment is critical. NeST offers a more efficient method for safety alignment compared to traditional fine-tuning, addressing the urgent need for adaptable and reliable safety mechanisms in AI systems.
Key Takeaways
- NeST selectively adapts a small subset of safety-relevant neurons to enhance model safety.
- It achieves a 90.2% reduction in unsafe outputs while requiring significantly fewer parameters compared to full fine-tuning.
- The framework enables rapid and stable safety updates without extensive model modification.
Computer Science > Cryptography and Security arXiv:2602.16835 (cs) [Submitted on 18 Feb 2026] Title:NeST: Neuron Selective Tuning for LLM Safety Authors:Sasha Behrouzi, Lichao Wu, Mohamadreza Rostami, Ahmad-Reza Sadeghi View a PDF of the paper titled NeST: Neuron Selective Tuning for LLM Safety, by Sasha Behrouzi and 3 other authors View PDF HTML (experimental) Abstract:Safety alignment is essential for the responsible deployment of large language models (LLMs). Yet, existing approaches often rely on heavyweight fine-tuning that is costly to update, audit, and maintain across model families. Full fine-tuning incurs substantial computational and storage overhead, while parameter-efficient methods such as LoRA trade efficiency for inconsistent safety gains and sensitivity to design choices. Safety intervention mechanisms such as circuit breakers reduce unsafe outputs without modifying model weights, but do not directly shape or preserve the internal representations that govern safety behavior. These limitations hinder rapid and reliable safety updates, particularly in settings where models evolve frequently or must adapt to new policies and domains. We present NeST, a lightweight, structure-aware safety alignment framework that strengthens refusal behavior by selectively adapting a small subset of safety-relevant neurons while freezing the remainder of the model. NeST aligns parameter updates with the internal organization of safety behavior by clustering functionally cohere...
