Re-Emergent Misalignment: How Narrow Fine-Tuning Erodes Safety Alignment in LLMs

TL;DR

This paper investigates how narrow fine-tuning of large language models on insecure code erodes their safety alignment by altering internal mechanisms, revealing the fragility of alignment and the need for more robust strategies.

Contribution

It provides a mechanistic analysis showing that narrow fine-tuning causes internal changes that oppose alignment, identifying shared latent dimensions affecting safety behaviors.

Findings

Fine-tuning on insecure code erodes prior alignment.

A shared latent dimension governs alignment behavior.

Narrow fine-tuning interferes with internal safety mechanisms.

Abstract

Recent work has shown that fine-tuning large language models (LLMs) on code with security vulnerabilities can result in misaligned and unsafe behaviors across broad domains. These results prompted concerns about the emergence of harmful behaviors from narrow domain fine-tuning. In this paper, we contextualize these findings by analyzing how such narrow adaptation impacts the internal mechanisms and behavioral manifestations of LLMs. Through a series of experiments covering output probability distributions, loss and gradient vector geometry, layer-wise activation dynamics, and activation space dimensions, we find that behaviors attributed to "emergent misalignment" may be better interpreted as an erosion of prior alignment. We show that fine tuning on insecure code induces internal changes that oppose alignment. Further, we identify a shared latent dimension in the model's activation space that governs alignment behavior. We show that this space is activated by insecure code and by misaligned responses more generally, revealing how narrow fine-tuning can degrade general safety behavior by interfering with shared internal mechanisms. Our findings offer a mechanistic interpretation for previously observed misalignment phenomena, and highlights the fragility of alignment in LLMs. The results underscore the need for more robust fine-tuning strategies that preserve intended behavior across domains.