Accidental Vulnerability: Factors in Fine-Tuning that Shift Model Safeguards

TL;DR

This paper investigates how fine-tuning large language models on specific datasets can unintentionally introduce vulnerabilities, analyzing dataset factors and proposing insights for better adversarial robustness and model safety.

Contribution

It identifies key dataset characteristics that contribute to accidental vulnerabilities in fine-tuned models and explores causal relationships to improve adversarial defenses.

Findings

Linguistic features, semantic similarity, and toxicity influence vulnerability levels.

Dataset characteristics correlate with attack success rates.

Understanding dataset factors aids in developing better defense strategies.

Abstract

As large language models (LLMs) gain popularity, their vulnerability to adversarial attacks emerges as a primary concern. While fine-tuning models on domain-specific datasets is often employed to improve model performance, it can inadvertently introduce vulnerabilities within the underlying model. In this work, we investigate Accidental Vulnerability, unexpected vulnerabilities arising from characteristics of fine-tuning data. We begin by identifying potential correlation factors such as linguistic features, semantic similarity, and toxicity across multiple experimental datasets. We then evaluate the adversarial robustness of these fine-tuned models, analyzing persona shifts and interpretability traits to understand how dataset factors contribute to attack success rates. Lastly, we explore causal relationships that offer new insights into adversarial defense strategies, highlighting the crucial role of dataset design in preserving model alignment. Our code is available at https://github.com/psyonp/accidental_vulnerability.