LLMTrace is a new large‑scale bilingual (English and Russian) corpus designed to improve AI‑written text detection. It contains character‑level annotations that enable precise localization of AI‑generated segments, supporting both full‑text binary classification and interval detection tasks. The dataset is built from a diverse set of modern proprietary and open‑source LLMs to address gaps in existing resources, such as outdated models, limited language coverage, and lack of mixed human‑AI authorship data.
By Irina Tolstykh, Aleksandra Tsybina, Sergey Yakubson, Maksim Kuprashevich
arXiv:2610.22940v1 Announce Type: new
Abstract: As large language models (LLMs) generate increasingly fluent text, the need for reliable detection of LLM-generated content has grown substantially. Ho...
By SungHo Kim, SangKeun Lee
arXiv:2609.22138v1 Announce Type: cross
Abstract: Hidden-state hallucination probing - training a linear classifier on an LLM's internal activations to detect whether a generated answer is faithful t...
By Tanveer Singh (Plaksha University)
IndicTriMix presents a new benchmark and models for token‑level language identification in tri‑language code‑mixed text involving Hindi, Gujarati, and Bengali. The authors reformulate the task as sequence labeling and fine‑tune transformer models MuRIL and XLM‑RoBERTa, evaluating them on manually annotated test sets. They also introduce two code‑mixing generation methods using parallel sentences and release the datasets and fine‑tuned models for public use.
By Pruthwik Mishra, Rudra Trivedi, Avi Patel, Ashok Urlana, Shrikant Malviya
arXiv:2606. 04177v1 Announce Type: cross Abstract: Interpretable linguistic features offer a promising approach for explaining why a given text appears machine-generated, particularly for non-expert users.
By Yassir El Attar, Esra D\"onmez, Maximilian Maurer, Agnieszka Falenska
VietAIDetector is an open‑source, zero‑shot tool for detecting Vietnamese AI‑generated text. It offers a Gradio web interface that accepts raw Vietnamese text, common file formats, scanned documents, and very long texts beyond typical LLM context limits. Built on a Vietnamese‑specific language model, it outperforms existing English‑centric methods on out‑of‑domain datasets and lets users choose detection thresholds based on F1, accuracy, or TPR@0.05FPR, with results viewable or downloadable as a PDF report.
By Trieu Hai Nguyen, Van-Dung Hoang
arXiv:2605.12890v2 Announce Type: replace-cross
Abstract: The rapid advancement of large language models (LLMs) has made machine-generated text increasingly difficult to distinguish from human-writte...
By Luxu Liang, Xiang Li
The paper titled "Safety Alignment Illusion: The Cross-Lingual Safety Gap in LLMs" highlights that current safety alignment training for large language models is predominantly English-centric, leading to failures in non‑English languages. It introduces INCLUDE, a multilingual benchmark with 2,604 prompts in six languages (English, Hindi, Bengali, Marathi, Tamil, and Hinglish) to measure Indian‑centric socio‑cultural biases. Evaluation of ten open‑ and closed‑source LLMs shows that Bengali models exhibit the highest bias scores among open‑source models, while English shows the lowest bias in open‑source but the highest in closed‑source models.
By Namya Bhatnagar
The paper introduces BanglaSafe, a benchmark of 879 Bengali prompts that covers 17 culturally grounded harm categories and five prompting conditions. Evaluation of 18 frontier LLMs shows that 53.6% of responses are unsafe or partially unsafe, with 14.7% containing strictly harmful content. The study finds that the writing style within Bengali has a stronger impact on safety than the language switch itself, and that current safety classifiers struggle to reliably evaluate Bengali content.
By Naymul Islam, Nusrat Jahan Lia, Shubhashis Roy Dipta, Sabik Bin Sultan, Abdullah Khan Zehady
arXiv:2610.16601v1 Announce Type: cross
Abstract: Processing multilingual text requires both identifying a broad range of languages and distinguishing the language of each token in inputs containing...
By Seongtae Hong, Youngjoon Jang, Seungyoon Lee, Jungseob Lee, Heuiseok Lim
arXiv:2606. 07706v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) have demonstrated strong performance across multimodal tasks, yet their safety robustness remains an open challenge.
By Rishabh Makwana, Mamta, Deeksha Varshney, Oana Cocarascu
IndicSafeEval is a new evaluation framework that tests the safety robustness of large language models against persuasion-based jailbreak attacks in Indian languages. The benchmark covers ten safety-critical content categories, six persuasive strategies, and four languages—Hindi, Bengali, Marathi, and Punjabi—producing 7,200 adversarial prompts. Experiments show that model safety varies significantly across languages, prompt styles, and risk categories, highlighting gaps in current English-centric safety assessments.