The paper investigates how newer GPT models alter scientific text, a phenomenon termed defensive writing, where the models either retract or weaken authors’ claims. Experiments show that as GPT versions advance, defensive writing becomes more pronounced, especially when the model is prompted to anticipate reviewer concerns. The study finds that this effect is largely driven by the anticipated-review explanation rather than mere correction of overclaiming, and that while AI reviewers rate such rewrites favorably, human readers perceive them as harder to read and less confident.
By Junchi Liao
Researchers find that newer GPT models increasingly rewrite academic paragraphs in a way that weakens or retracts the original authors’ claims—a phenomenon they term "defensive writing." The study shows that this effect grows with GPT version, with GPT‑6‑astra often removing claims outright and adding ungrounded qualifiers when rewriting from evidence sheets. The authors suggest that the models are likely tailoring text for an anticipated reviewer rather than simply correcting overclaiming, and that this defensive style can be amplified when AI writing is combined with AI review.
The paper introduces a harness designed to curb over‑defensive writing by agents, which occurs when agents add stronger claims than the source material and then retract them. The harness separates the reviewer, filter, and writer components, allowing the writer to focus only on problems selected by an independent filter. Evaluation on 25 business‑project drafts shows a 96.61% drop in defensive propositions and an increase in source‑claim retention from 80.35% to 93.12%.
By Xiaochen Chang
arXiv:2608. 08975v1 Announce Type: cross Abstract: As large language models increasingly participate in scientific evaluation, we investigate a potential form of reward hacking: how rhetorical choices shape AI-review judgments when reported scientific content is preserved and how these effects vary across evaluation conditions.
By Ming Li, Chenguang Wang, Xirui Li, Xinyue Zeng, Dianqi Li, Peng Shi, Dawei Zhou, Tianyi Zhou
The study examines how the rapid turnover of large language models (LLMs) undermines the effectiveness of AI‑detectors used to screen scientific manuscripts. By comparing 4,000 PNAS abstracts rewritten by 23 different LLM versions released between June 2023 and August 2026, the authors show that detectors trained on past versions can fail dramatically at generation boundaries, missing up to 96% of rewrites just after a sharp transition. Even detectors updated to newer versions can overlook rewrites of older models, and a commercial detector missed most rewrites of the newest version while flagging few human‑written abstracts.
By Kazuki Nakajima, Takayuki Mizuno
The paper introduces Thinking‑in‑Bets, a training‑free technique where a language‑model agent records its key claims, supporting checks, and confidence before receiving user feedback. By consulting this private claim checklist when challenged, the agent preserves verified work and focuses on uncertain or newly contested parts, reducing sycophantic revisions. Experiments on five Gemini models across eight benchmarks show that pre‑reply claim records nearly eliminate false reversals in patch review, protect report drafts from mistaken pushback, and lower unnecessary edits in multi‑turn coding while improving accuracy and confidence in non‑interactive tasks.
By Nghi D. Q. Bui
arXiv:2605. 04135v2 Announce Type: replace-cross Abstract: Readers of applied-domain LLM capability evaluations want to know what AI systems can currently do.
By David Gringras, Misha Salahshoor
The paper introduces EquiReview‑R, an AI‑assisted review system that treats omission and over‑critique as distinct risks and refines a structured concern set using evidence‑linked reasoning. It demonstrates that more criticism does not guarantee a better review, showing that many high‑recall reviews lack definitive evidence for concerns and that revision before further search is essential. On a held‑out set of papers, EquiReview‑R meets non‑inferiority for major omission, cuts major over‑critique from 15.5 % to 8.1 %, and stops on 52.4 % of papers, with gains attributed to revision rather than extra inference.
By Zexing Zhang, Jichao Li, Tianyang Lei, Yude Fu, Yang Kewei
As large language models increasingly participate in scientific evaluation, we investigate a potential form of reward hacking: how rhetorical choices shape AI-review judgments when reported scientific content is preserved and how these effects vary across evaluation conditions. We construct a controlled corpus of 4,200 full-paper manuscripts derived from 120 anonymized ICLR 2026 submissions.
arXiv:2610.10092v1 Announce Type: new
Abstract: For better or worse, LLMs are by now used routinely for scientific writing.\footnote{This paper is no exception; we did use AI to assist with writing s...
By Olga Zamaraeva, Adri\'an Gude, Roi Santos-R\'ios, Carlos G\'omez-Rodr\'iguez
The paper investigates how mainstream document converters lose formatting cues—specifically highlights and colours—when turning Word or PDF files into tokens for enterprise agents. Using a dataset of 76 policy documents, the authors show that this loss dramatically degrades agents’ ability to recover author‑specified requirements, dropping micro‑F1 scores from 0.690 to as low as 0.075. The study also demonstrates that explicitly adding cues back into the text can partially recover performance and that the issue is measurable and inexpensive to fix.
By Yikai Zhao, Saurabh Pandey, Pradeep Kumar Misra
arXiv:2604.25213v2 Announce Type: replace
Abstract: OpenAI's GPT-Image-2 has effectively erased the visual boundary between authentic and AI-edited document images: a single number on a receipt can b...
By Jiaqi Wu, Yuchen Zhou, Dennis Tsang Ng, Xingyu Shen, Kidus Zewde, Ankit Raj, Tommy Duong, Simiao Ren