Jun 2026· arXiv.org· Vol abs/2606.31074· 0 citations· 74 references
Computer Science
TL;DR
This study proposes a novel Triospect Detection Framework by using additional perspectives of content and expression within a given text to enhance detection reliability against attacks.
Abstract
Existing AI-generated text detectors are vulnerable to attacks that manipulate textual characteristics. In this study, we propose a novel Triospect Detection Framework by using additional perspectives of content (core ideas) and expression (stylistic elements) within a given text. Experiments on two benchmarks involving 17 attacks, 12 domains, and 17 source models demonstrate that Triospect is robust against these attacks. It improves the strong baseline by a significant margin of 22.3% (AUROC) and 13% (TPR01) on the Humanize-16K after-attack subset, and by 9.1% (AUROC) and 22% (TPR01) on the adversarial RAID. This framework marks a pioneering effort in statistical methods to enhance detection reliability against attacks. We release our data and code at https://github.com/baoguangsheng/triospect.
Large language models (LLMs) have become capable of producing human-like prose, institutions ranging from universities to publishers have adopted automated AI-text detectors — most visibly Turnitin's AI writing indicator, GPTZero, and similar tools — as a control against misrepresenting machine-generated content as human work. This paper examines how these detectors work, the empirical and theoretical evidence on their reliability, and the documented techniques that allow AI-generated text to evade them. Drawing on the peer-reviewed and preprint literature, we describe three detector families (zero-shot statistical detectors, trained classifiers, and watermarking schemes), summarize adversarial results showing that paraphrasing attacks can collapse watermark detection true-positive rates from above 99% to under 10%, and review evidence that current detectors produce systematically higher false positive rates for non native English writers — in one widely cited study, 61.3% of TOEFL essays were misclassified as AI generated versus near zero misclassification of native-speaker essays. We conclude that AI-text detection, in its current form, cannot serve as a sole, dispositive basis for academic-integrity decisions, and we propose a set of institutional and technical recommendations — process-based evidence, disclosed AI-use policies, watermarking at the model level, and human-adjudicated review — that better address the underlying problem than detector accuracy alone can.
A. Kumar, Ayush Kumar, Danish Maqbool Chopan· International Journal of Res...· 0 citations
A hybrid detection framework which combines semantically deep embeddings from the RoBERTa transformer with a set of carefully designed language statistics and linguistic statistics and shows excellent resistance to the surface-level adversarial paraphrasing strategy.
Anita Rani, Suman· International Journal of Sci...· 0 citations
This paper explores the application of natural language processing and machine learning techniques to automatically detect and classify fake news articles and develops a web-based application to integrate the trained models into an interactive platform.
Vlad-Constantin Cristescu, Stelian Spinu· Journal of Military Technolo...· 0 citations
Phishing remains a prevalent cyberattack vector, and the widespread adoption of large language models (LLMs) has enabled adversaries to generate grammatically correct and contextually coherent phishing emails at scale, against which conventional detection systems are less effective. Although stylometric methods achieve over 95% accuracy within a single generator, their performance has not been systematically evaluated when the source model changes between training and deployment. This represents a significant gap, as adversaries can switch generators rapidly. A balanced corpus of 9,986 phishing emails was assembled, comprising 4,986 emails generated by three modern LLMs (GPT-4.1, DeepSeek 3.2, and Llama 3.3 70B) across five thematic categories, and 5,000 human phishing emails sampled in a stratified manner from five public sources. Seventeen stylometric features were extracted, and Logistic Regression and XGBoost classifiers were evaluated under intra-model, cross-model, threshold-recalibrated, cross-dataset, and aggregated-pool settings. Intra-model F1 scores reached 0.96 under stratified cross-validation and 0.999 on held-out splits used for the cross-model matrix. However, cross-model F1 dropped by 28.0 percentage points under the default decision threshold of 0.5. Notably, the area under the receiver operating characteristic curve remained above 0.96 in every off-diagonal cell, indicating that discriminative information is preserved even though the decision threshold is generator-specific. Recalibrating the threshold on a small target subset reduced the gap to 4.0 percentage points (an 86% reduction), and an aggregated-pool detector achieved F1 = 0.997 on each generator. This work reframes cross-model phishing detection from a problem of model incompatibility to one of practical calibration, and provides two deployable solutions, threshold recalibration on a small target slice and aggregated-pool training, along with a publicly released multi-LLM corpus.
Rommel Gutierrez, W. Villegas-Ch., Jaime Govea· Frontiers in Big Data· 0 citations
The core of VaryBalance is that, compared to LLM-generated texts, there is a greater difference between human texts and their rewritten version via LLMs, and quantifies this through Mean Squared Deviation and distinguishes human texts and LLM-generated texts.
Xuecong Li, Xiaohong Li, Qiang Hu et al.· 0 citations
Large Language Models (LLMs) are now widely used to draft, revise, paraphrase, and polish text, making the detection of AI-generated writing increasingly difficult. This systematic literature review synthesizes peer-reviewed and high-quality studies published between 2023 and 2026 on AI-obfuscated, AI-refined, and humanized text. From 1,002 records, 26 primary studies were retained after screening and quality assessment. The review organizes the literature through a seven-dimensional taxonomy. Overall, the evidence shows that many detectors perform well on clean or in-distribution AI-text but become less reliable when the text is paraphrased, humanized, or collaboratively edited. The review also highlights recurring fairness concerns, especially for non-native English writers, and finds that current benchmarks often do not fully capture realistic mixed-authorship and adversarial settings. These results suggest that AI-text detection should be treated as one supportive signal rather than a stand-alone judgment, particularly in high-stakes academic or professional contexts.
Batyr Sharimbayev, S. Kadyrov· Engineering, Technology &...· 0 citations