Stress-testing Machine Generated Text Detection: Shifting Language Models Writing Style to Fool Detectors

TL;DR

This paper evaluates the robustness of machine-generated text detectors against adversarial attacks that shift generated text style towards human writing, revealing detectors' vulnerability and the need for more resilient detection methods.

Contribution

It introduces a pipeline to test detector resilience by fine-tuning language models to mimic human writing, exposing weaknesses in current detection approaches.

Findings

Detectors are easily fooled with few examples.

Detection performance drops significantly under adversarial style shifts.

Linguistic features exploited by detectors are identified.

Abstract

Recent advancements in Generative AI and Large Language Models (LLMs) have enabled the creation of highly realistic synthetic content, raising concerns about the potential for malicious use, such as misinformation and manipulation. Moreover, detecting Machine-Generated Text (MGT) remains challenging due to the lack of robust benchmarks that assess generalization to real-world scenarios. In this work, we present a pipeline to test the resilience of state-of-the-art MGT detectors (e.g., Mage, Radar, LLM-DetectAIve) to linguistically informed adversarial attacks. To challenge the detectors, we fine-tune language models using Direct Preference Optimization (DPO) to shift the MGT style toward human-written text (HWT). This exploits the detectors' reliance on stylistic clues, making new generations more challenging to detect. Additionally, we analyze the linguistic shifts induced by the alignment and which features are used by detectors to detect MGT texts. Our results show that detectors can be easily fooled with relatively few examples, resulting in a significant drop in detection performance. This highlights the importance of improving detection methods and making them robust to unseen in-domain texts.