Refusal Behavior in Large Language Models: A Nonlinear Perspective

Fabian Hildebrandt, Andreas Maier, Patrick Krauss, Achim Schilling
1/14/2025
cs.CLcs.AI

No official code found. Best implementation: FabianHildebrandt/Refusal-LLMs (3 stars, MIT, last commit 2025-01-30).

Abstract

Refusal behavior in large language models (LLMs) enables them to decline responding to harmful, unethical, or inappropriate prompts, ensuring alignment with ethical standards. This paper investigates refusal behavior across six LLMs from three architectural families. We challenge the assumption of refusal as a linear phenomenon by employing dimensionality reduction techniques, including PCA, t-SNE, and UMAP. Our results reveal that refusal mechanisms exhibit nonlinear, multidimensional characteristics that vary by model architecture and layer. These findings highlight the need for nonlinear interpretability to improve alignment research and inform safer AI deployment strategies.

View on arXivView PDF

Code Implementations(1)

FabianHildebrandt/Refusal-LLMsCommunity100%
MIT
30Python, Jupyter NotebookJan 11, 20251 years agoMIT

Similar Papers

The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence

89% match1 repo362025

There Is More to Refusal in Large Language Models than a Single Direction

89% match2026

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models

88% match1 repo112024

Understanding Refusal in Language Models with Sparse Autoencoders

88% match1 repo102025

OR-Bench: An Over-Refusal Benchmark for Large Language Models

88% match1 repo282024

OR-Bench: An Over-Refusal Benchmark for Large Language Models

88% match1 repo282026

Cite this paper

@article{hildebrandt2025refusal,
  title  = {Refusal Behavior in Large Language Models: A Nonlinear Perspective},
  author = {Fabian Hildebrandt and Andreas Maier and Patrick Krauss and Achim Schilling},
  year   = {2025},
  eprint = {2501.08145},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2501.08145v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2501.08145.svg)](https://codeofpaper.com/paper/2501.08145)

Discussion