Attacks on Machine-Text Detectors Retain Stylistic Fingerprints
Abstract
Machine-text detection remains challenging despite evasion techniques, but stylistic features can provide robust defense when analyzed across multiple documents rather than individual instances.
Despite considerable progress in the development of machine-text detectors, the ease with which machine-text can be manipulated to evade detection has led to suggestions that the problem is inherently intractable. In this work, we investigate the limits of such evasion strategies. We demonstrate that while current attacks, ranging from prompt engineering to detector-guided optimization can effectively degrade performance of standard detectors, they fail to erase the underlying stylistic "fingerprints" of machine text. We show that few-shot detectors that utilize the stylistic feature space are robust to these evasion attempts, reliably detecting samples even from models explicitly tuned to prevent detection. This raises the question: does style represent a universal defense against machine-detection attacks? We demonstrate that the answer is "no'' by introducing a novel paraphrasing approach that simultaneously optimizes for undetectability and adherence to specific human styles. We show that unlike prior methods, this attack effectively evades all considered detectors, including those that utilize writing style. However, we find that this evasion is not absolute: as the number of documents available for analysis grows, the human and machine distributions become distinguishable again. Overall, our findings suggest that reliable machine-text detection requires moving beyond single-document analysis to multi-document analysis.
Community
Existing evasion attacks can fool standard machine-text detectors, but they do not remove the stylistic fingerprint of machine-generated text. As a result, detectors that leverage style remain robust. We show that it is possible to construct a style-aware paraphrasing attack that jointly optimizes for undetectability and alignment with a target author’s style, evading all detectors when detection relies on a single document. However, when multiple documents are aggregated, the human and machine distributions separate again.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Unsupervised Style Representation Learning for AI-Text Detection via Paraphrase Inversion (2026)
- Non-Parametric Machine Text Detection via Multi-View Gaussian Processes (2026)
- Fight Poison with Poison: Enhancing Robustness in Few-shot Machine-Generated Text Detection with Adversarial Training (2026)
- SV-Detect: AI-generated Text Detection with Steering Vectors (2026)
- Base Models Look Human To AI Detectors (2026)
- DSIPA: Detecting LLM-Generated Texts via Sentiment-Invariant Patterns Divergence Analysis (2026)
- LLM Output Detectability and Task Performance Can be Jointly Optimized (2026)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend
Get this paper in your agent:
hf papers read 2505.14608 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 1
Datasets citing this paper 2
rrivera1849/style-aware-paraphraser-author-bank-reddit
rrivera1849/style-aware-paraphraser-outputs
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper