Investigating the Influence of Prompt-Specific Shortcuts in AI Generated Text Detection

TL;DR

This paper investigates how prompt-specific shortcuts affect AI-generated text detection, revealing vulnerabilities and proposing a method to improve detector robustness through adversarial prompt optimization.

Contribution

It introduces FAILOpt, an adversarial attack that exploits prompt shortcuts, and demonstrates how to use it to enhance the robustness of AI-generated text detectors.

Findings

FAILOpt effectively reduces detector performance.

Augmented training improves detection across models and tasks.

Prompt shortcuts significantly impact detection robustness.

Abstract

AI Generated Text (AIGT) detectors are developed with texts from humans and LLMs of common tasks. Despite the diversity of plausible prompt choices, these datasets are generally constructed with a limited number of prompts. The lack of prompt variation can introduce prompt-specific shortcut features that exist in data collected with the chosen prompt, but do not generalize to others. In this paper, we analyze the impact of such shortcuts in AIGT detection. We propose Feedback-based Adversarial Instruction List Optimization (FAILOpt), an attack that searches for instructions deceptive to AIGT detectors exploiting prompt-specific shortcuts. FAILOpt effectively drops the detection performance of the target detector, comparable to other attacks based on adversarial in-context examples. We also utilize our method to enhance the robustness of the detector by mitigating the shortcuts. Based on the findings, we further train the classifier with the dataset augmented by FAILOpt prompt. The augmented classifier exhibits improvements across generation models, tasks, and attacks. Our code will be available at https://github.com/zxcvvxcz/FAILOpt.