The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; 16 of the 20 sources its claims cite were checked for support.
Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text
Open accessAbstract
This work introduces Adversarial Paraphrasing, a training-free attack framework that universally humanizes any AI-generated text to evade detection more effectively. Our approach leverages an off-the-shelf instruction-following LLM to paraphrase AI-generated content under the guidance of an AI text detector, producing adversarial examples that are specifically optimized to bypass detection. Extensive experiments show that our attack is both broadly effective and highly transferable across several detection systems. For instance, compared to simple paraphrasing attack--which, ironically, increases the true positive at 1% false positive (T@1%F) by 8.57% on RADAR and 15.03% on Fast-DetectGPT--adversarial paraphrasing, guided by OpenAI-RoBERTa-Large, reduces T@1%F by 64.49% on RADAR and a striking 98.96% on Fast-DetectGPT. Across a diverse set of detectors--including neural network-based, watermark-based, and zero-shot approaches--our attack achieves an average T@1%F reduction of 87.88% under the guidance of OpenAI-RoBERTa-Large. We also analyze the tradeoff between text quality and attack success, and show that our method preserves the high quality of the generated text. Our findings highlight the need for more robust and resilient detection strategies in the light of increasingly sophisticated evasion techniques.
Overview
CoChat Summary
TL;DR
The paper introduces Adversarial Paraphrasing, a training-free attack that uses an instruction-tuned LLM to paraphrase AI-generated text while a detector guides each token choice toward lower AI scores. It reports broad evasion and transferability across neural, watermark, and zero-shot detectors, with mostly slight text-quality degradation.
Key findings
- Guided by OpenAI-RoBERTa-Large, adversarial paraphrasing reduces T@1%F by 64.49% on RADAR and 98.96% on Fast-DetectGPT, whereas simple paraphrasing increases T@1%F by 8.57% on RADAR and 15.03% on Fast-DetectGPT.
- Across eight detectors, adversarial paraphrasing achieves an average T@1%F reduction of 87.88% under OpenAI-RoBERTa-Large guidance; average relative drops are 84.94% with MAGE guidance, 86.89% with RADAR, 80.75% with OpenAI-RoBERTa-Base, and 87.88% with OpenAI-RoBERTa-Large, compared with 30.27% for simple paraphrasing.
- Text quality is largely preserved: across datasets and guidance detectors, 87% of adversarially paraphrased texts received GPT-4o ratings of 4 or 5; SBERT cosine similarity was slightly lower than simple paraphrasing (e.g., 0.8082 for RoBERTa-Large guidance vs. 0.8601 for simple), and perplexity was comparable to human text (AdvPara RoBERTa-Large 14.26 ± 4.97 vs. human 15.02 ± 7.71 and simple 9.28 ± 3.86).
- Per-sample runtime was 7.18 ± 0.13 s for simple paraphrasing and 10.20 ± 0.18 s for AdvPara RoBERTa-Large, 8.64 ± 0.11 s for RoBERTa-Base, 16.71 ± 0.74 s for MAGE, and 9.69 ± 0.20 s for RADAR.
Methods
- The attack is training-free and gradient-free: LLaMA-3-8B-Instruct is configured as a paraphraser via a system prompt, and at each auto-regressive step it applies top-p = 0.99 and top-k = 50 filtering, scores candidate continuations with a guidance AI text detector, and selects the token with the lowest AI score.
- Evaluation uses eight detectors: neural network-based (OpenAI-RoBERTa-Base, OpenAI-RoBERTa-Large, MAGE, RADAR), watermark-based (KGW, Unigram), and zero-shot (Fast-DetectGPT, GLTR). For non-watermark detectors, the main dataset is MAGE with 2,000 AI-generated and 2,000 human-written texts of about 100–500 tokens; watermark datasets are generated with watermarked LLaMA-3.1-8B-Instruct from first 20-word prefixes, about 200–600 tokens.
- Main metrics are AUC, T@1%F, and GPT-4o quality ratings; baselines are simple paraphrasing and recursive paraphrasing, with watermark stealing compared in Appendix A. Additional analyses cover transferability, efficiency, top-p/top-k ablations, guidance-frequency ablations, and text quality via perplexity, SBERT similarity, GPT-4o ratings, and win rates.
Limitations
- The paper states the attack is not always successful: in 24,000 samples under a fixed 0.5 threshold in transfer settings, it failed in 1,773 cases (~7.4%) and backfired in 995 cases (~4.15%).
- The paper states a slight tradeoff in text quality and higher latency than simple paraphrasing; it notes the quality difference is not statistically significant and that MAGE guidance has higher latency due to its LongFormer-based architecture.
- Evident from the text: evaluation is limited to MAGE and watermarked LLaMA-3.1 datasets, eight detectors, English text, and automated GPT-4o quality judgments rather than human evaluation.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
12 Valid · 3 Not valid · 4 Not checked · 68 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“However, adversarial paraphrasing significantly reduces RADAR's detection.”
What the sources say
- Disputedcites [8] [8] X. Hu, P.-Y. Chen, and T.-Y. Ho. Radar: Robust ai-text detection via adversarial learning. Advances in neural information processing systems , 36:15077-15095, 2023. - sourceExperimental — accuracy not yet measured
The Introduction states that RADAR's detector is not weakened by paraphrasing. Section 4.2 says its detection performance is stable across paraphrasing schemes and that RADAR mitigates the performance drop seen in other methods. These passages contradict the claim that adversarial paraphrasing significantly reduces RADAR's detection. (LLM, conf=0.90)
Checked by a reasoning model — may be wrong.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
Every source the paper cites for this claim was checked and supports it.
In the paper
“Notably, RADAR [8]-a detector adversarially trained to be robust to paraphrasing attacks-exhibits improved detection rates after baseline simple and recursive paraphrasing attacks.”
What the sources say
- Supportedcites [8] [8] X. Hu, P.-Y. Chen, and T.-Y. Ho. Radar: Robust ai-text detection via adversarial learning. Advances in neural information processing systems , 36:15077-15095, 2023. - source
Section 4.2 reports that RADAR outperforms the compared detectors under paraphrasing and maintains stable detection performance across paraphrasing schemes. Section 4.4 reports that RADAR remains robust when texts are paraphrased multiple times. The paper reports AUROC rather than a detection rate, but the claimed improvement relative to baseline detectors is supported. (LLM, conf=0.87)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The results indicate that adversarial paraphrasing guided by RADAR yields the lowest AUC and T@1%F values post-attack.”
What the sources say
- Partly supportedcites [8] [8] X. Hu, P.-Y. Chen, and T.-Y. Ho. Radar: Robust ai-text detection via adversarial learning. Advances in neural information processing systems , 36:15077-15095, 2023. - sourceExperimental — accuracy not yet measured
Section 4.2 reports that paraphrasing substantially lowers the AUROC of most existing detectors, supporting the general post-attack degradation. It does not establish that RADAR-guided paraphrasing yields the lowest AUC across attacks, and the paper does not report T@1%F. Section 4.2 instead says RADAR's detector performs best under paraphrasing. (LLM, conf=0.85)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“On average, simple paraphrasing leads to a 30.27% relative drop in T@1%F.”
What the sources say
- Partly supportedcites [15] [15] K. Krishna, Y. Song, M. Karpinska, J. Wieting, and M. Iyyer. Paraphrasing evades detectors of ai-generated text, but retrieval is an effective defense. Advances in Neural Information Processing Systems , 36:27469-27500, 2023. - sourceExperimental — accuracy not yet measured
Section 4.1 defines detection accuracy at a 1% false-positive rate, and Section 4.3 reports that paraphrasing significantly lowers it. But the paper reports results for DIPPER paraphrasing and does not state an average relative drop of 30.27% or identify that figure with simple paraphrasing. (LLM, conf=0.87)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Notably, RADAR [8] proves to be the most effective guidance detector in this setting.”
What the sources say
- Partly supportedcites [8] [8] X. Hu, P.-Y. Chen, and T.-Y. Ho. Radar: Robust ai-text detection via adversarial learning. Advances in neural information processing systems , 36:15077-15095, 2023. - sourceExperimental — accuracy not yet measured
The Introduction reports that RADAR outperformed six existing AI-text detectors when detecting text from an unseen paraphraser. It does not evaluate RADAR as a guidance detector in a watermark-stealing comparison, so it cannot establish that RADAR was the most effective detector in that setting. (LLM, conf=0.82)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
51 more claims
Create a free account to see all 87 claims and every source verdict.
- Probably not a substantial claim. 10
- It may state two separate claims as one. 23
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.1 Setup
“We use LLaMA-3-8B-Instruct [21] with a custom system prompt (see Figure 2) as our paraphraser model.”
§ 5 Quality Evaluation of Paraphrased Texts
“This may be attributed to the fact that the model used for paraphrasing (LLaMA-3.1) is superior to the LLMs used for generating AI texts in the MAGE dataset”
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 2
- [17][17] T. Lavergne, T. Urvoy, and F. Yvon. Detecting fake content with relative entropy scoring. Pan , 8(27-31):4, 2008. -
- [28][28] V. S. Sadasivan, A. Kumar, S. Balasubramanian, W. Wang, and S. Feizi. Can AI-generated text be reliably detected? stress testing AI text detectors under various attacks. Transactions on Machine Learning Research , 2025. -
Cited by
ToolTweak: An Attack on Tool Selection in LLM-based AgentsPreprint
Sneh, J., Yan, R., Yu, J., Torr, P. H. S., Gal, Y., Sengupta, S., Sommerlade, E., Paren, A., & Bibi, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.02554 · Cited by 47
Modeling the Attack: Detecting AI-Generated Text by Quantifying Adversarial Perturbations
Teja, L. D. M. S. S., Yadagiri, A., Anish, S. S., Nuthakki, S. G. K., & Pakray, P. (2026) · International Conference on Ubiquitous Information Management and Communication (IMCOM) · doi:10.1109/imcom69009.2026.11360855 · Cited by 9
Hu, W., Li, H., Jing, H., Hu, Q., Zeng, Z., Han, S., Heli, X., Chu, T., Hu, P., & Song, Y. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.44 · Cited by 16
On the Detectability of LLM-Generated Text: What Exactly Is LLM-Generated Text?Preprint
Geng, M., & Poibeau, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.20810 · Cited by 6
EditLens: Quantifying the Extent of AI Editing in Text
Thai, K., Emi, B., Masrour, E., & Iyyer, M. (2025) · arXiv · doi:10.48550/arxiv.2510.03154 · Cited by 3
Metadata
- Authors
- Cheng, Y., Sadasivan, V. S., Saberi, M., Saha, S., & Feizi, S.
- Affiliations
- University of Maryland, College Park (US)
- UMD CP
- Venue
- Advances in Neural Information Processing Systems 38
- Published
- 2025
- Pages
- 53123-53154
- ISSN
- 1049-5258 (ISSN-L)
- OpenAlex
- W4417126423
- Cited by
- 3
- References
- 0
- Citations by year
- 2026: 3
- FWCI
- 2.82
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- neural information processing systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Authorship Attribution and Profiling
- Keywords
- AI-generated text detection, evasion attacks, large language models, social engineering, adversarial examples, zero-shot detection
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (prompt extract-v3).
A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text. https://app.cochat.ai/papers/10.52202/085713-1588
- The paper
- Cheng, Y., Sadasivan, V. S., Saberi, M., Saha, S., & Feizi, S. (2025). Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text. Advances in Neural Information Processing Systems 38. https://doi.org/10.52202/085713-1588
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.