StealthRL：面向多检测器规避的强化学习复述攻击研究

View PDF HTML (experimental)

Abstract:AI-text detectors are increasingly used in high-stakes settings, but their robustness to meaning-preserving adversarial rewriting remains unclear. We introduce StealthRL, a reinforcement learning framework for stress-testing detectors with adaptive paraphrase attacks. StealthRL trains a paraphrase policy against a detector ensemble while enforcing semantic fidelity, then evaluates transfer to held-out detector families on the full filtered MAGE test pool. Across RoBERTa, Fast-DetectGPT, Binoculars, and MAGE, the resulting attack sharply degrades detection performance, driving mean TPR@1%FPR down to 0.024 while retaining nontrivial semantic overlap and readable prose. The attack also transfers to detectors not used during training, indicating a broader robustness failure rather than overfitting to a single target model. We complement the main results with detector-score analysis, repeated-generation stability checks, and automatic and judge-based quality evaluation. Our findings suggest that current AI-text detectors remain brittle under adaptive paraphrasing pressure and motivate robustness evaluation at deployment-relevant operating points. Our code and the trained model checkpoint are publicly available.

Comments:	Long paper at the First Workshop on Pluralistic Value Alignment of LLMs (PlurVA-LLM 2026) @ AACL-IJCNLP 2026. 16 pages. Code: this https URL
Subjects:	Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
Cite as:	arXiv:2602.08934 [cs.LG]
	(or arXiv:2602.08934v3 [cs.LG] for this version)
	https://doi.org/10.48550/arXiv.2602.08934 arXiv-issued DOI via DataCite

Submission history

From: Suraj Ranganath [view email]
[v1] Mon, 9 Feb 2026 17:33:46 UTC (13,020 KB)
[v2] Thu, 19 Mar 2026 23:19:31 UTC (8,988 KB)
[v3] Fri, 23 Oct 2026 08:21:02 UTC (755 KB)