AUROC gains that looked solid on two corpora and nine AI-text detectors disappear under a two-sided protocol, where attacks are applied to both human and machine text; in the same experiments, attacked human text is misclassified at rates up to 100%. Another paper compares text-only and image-aware vision-language model judges for manga translation evaluation, using page images to capture speaker identity, referents and scene tone that the bubble text alone leaves out.

Attacking Only Half the Data: Two-Sided Evaluation of AI-Text Detector Robustness

If you use AI-text detectors in translation or QA workflows, score them on both human and machine text, because two-sided testing wipes out the reported AUROC gains and attacks can push human text to 100% false positives.

We investigate the robustness of widely used AI text detectors under attacks applied to both human and machine text. Prior evaluations typically attack only machine text, overlooking that human text is routinely modified through grammar correction, formatting, or invisible character insertion. We introduce a two-sided protocol that applies attacks to both classes. Across two corpora and nine detectors, reported AUROC gains vanish under two-sided evaluation, with attacked human text misclassified at rates up to 100%. While simple preprocessing reverses character-level attacks, word-level synonym substitution remains effective meaning-preserving, and irreversible.

Text-Only vs. Image-Aware VLM Judges for Manga Translation Evaluation

For manga localisation, include page images in automatic evaluation when visual context matters, because the same VLM catches more visually dependent errors, though human checks still do not favour text-only or image-aware judging overall.

In manga translation, visual context outside speech-bubble text, including speaker identities, referents, and scene tone, is depicted in page images, making it natural to refer to images during translation evaluation. However, most conventional manga translation evaluations rely solely on text-only automatic metrics, leaving the impact of image input insufficiently explored. We propose a practical validation framework that employs the same vision-language model (VLM) as both a text-only judge (TOJ) and an image-aware judge (IAJ) to investigate evaluation divergences associated with image-aware evaluation procedures. We use a Japanese-English manga dataset of official translations aligned at the page and speech-bubble levels. First, adding page images to TOJ improved the primary VLM's detection of deliberately injected, visually dependent degradations. Second, we analyzed evaluation divergences between TOJ and IAJ, finding different error counts and error-category distributions. Human verification did not reveal a consistent preference between TOJ and IAJ.

Papers announced by arXiv on Tuesday, 13 October 2026, covering submissions from 12 October to 13 October 2026.