Paraphrasing Attack Resilience of Various AI-Generated Text Detection Methods
For researchers and practitioners deploying AI text detectors, this work highlights that high-performing methods may be brittle under paraphrasing attacks, complicating reliability assessments.
The paper evaluates three AI-generated text detection methods (fine-tuned RoBERTa, Binoculars, and text feature analysis) and their ensembles against paraphrasing attacks, finding that Binoculars-inclusive ensembles achieve the highest performance but suffer the largest drops under attack, revealing a trade-off between performance and resilience.
The recent large-scale emergence of LLMs has left an open space for dealing with their consequences, such as plagiarism or the spread of false information on the Internet. Coupling this with the rise of AI detector bypassing tools, reliable machine-generated text detection is in increasingly high demand. We investigate the paraphrasing attack resilience of various machine-generated text detection methods, evaluating three approaches: fine-tuned RoBERTa, Binoculars, and text feature analysis, along with their ensembles using Random Forest classifiers. We discovered that Binoculars-inclusive ensembles yield the strongest results, but they also suffer the most significant losses during attacks. In this paper, we present the dichotomy of performance versus resilience in the world of AI text detection, which complicates the current perception of reliability among state-of-the-art techniques.