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

2026-02-02

AI Summary

This study investigates whether refusal behaviors in large language models are governed by a single activation direction. Through geometric analysis of the activation space and linear intervention techniques, the authors systematically examine the representational structure underlying eleven categories of refusal and noncompliant behaviors. The findings reveal that these behaviors correspond to multiple distinct activation directions, indicating directional diversity in refusal mechanisms. Nevertheless, manipulating any one of these directions produces a similar trade-off between appropriate refusal and over-refusal, suggesting the presence of a shared one-dimensional control mechanism. This work is the first to demonstrate that refusal behavior is co-regulated by multiple directions, with different directions primarily modulating how refusal is expressed rather than whether it occurs, thereby significantly advancing the understanding of model safety mechanisms.

Technology Category

• Natural Language Processing: Safety and Robustness
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Responsible Web: Social and technical mechanisms of refusal for web technologies and applications
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Prior work argues that refusal in large language models is mediated by a single activation-space direction, enabling effective steering and ablation. We show that this account is incomplete. Across eleven categories of refusal and non-compliance, including safety, incomplete or unsupported requests, anthropomorphization, and over-refusal, we find that these refusal behaviors correspond to geometrically distinct directions in activation space. Yet despite this diversity, linear steering along any refusal-related direction produces nearly identical refusal to over-refusal trade-offs, acting as a shared one-dimensional control knob. The primary effect of different directions is not whether the model refuses, but how it refuses.

Problem

• refusal
• large language models
• activation space
• steering
• non-compliance

Innovation

• refusal directions
• activation space
• linear steering
• large language models
• behavioral diversity