Large language model changes weaken ai writing detection in science
Large Language Model Turnover Undermines Screening for Artificial Intelligence-Assisted Scientific Writing
Computation and LanguageArtificial IntelligenceComputers and SocietyDigital LibrariesSocial and Information Networks
Summary
Detecting scientific papers partly written by AI relies on recognizing patterns from specific AI versions. The authors show that as AI language models get updated, detectors trained on older versions struggle to catch new AI writing or may falsely flag normal human writing. They tested 23 versions of AI models rewriting scientific abstracts and found that detectors can miss many AI rewrites or wrongly accuse humans depending on how often the detector updates. This means checking for AI use in papers must be updated regularly to stay accurate.
What this means in practice
- •For journal editors: Adjust manuscript screening protocols to update AI detectors regularly to maintain detection accuracy across LLM versions.
- •For conference organizers: Improve submission review processes by verifying AI detection tools work well with the latest LLM versions to avoid missing AI-written content.
Authors
Kazuki Nakajima, Takayuki Mizuno
Abstract
Journals and conferences have begun to screen submitted manuscripts for text written using large language models (LLMs). The reliability of this screening rests on benchmark evaluations against a fixed set of LLM versions, while the versions in actual use keep changing. Here we quantify how this LLM turnover affects the screening of scientific manuscripts. We paired 4,000 pre-ChatGPT abstracts from the Proceedings of the National Academy of Sciences with their rewrites by 23 LLM versions from three vendors, released between June 2023 and August 2026. We then trained detectors under maintenance scenarios ranging from a detector retrained on every new version to one trained once and never updated. Detectors trained only on a vendor's past versions can collapse at the boundaries between model generations: calibrated to falsely flag 1% of human-written abstracts, they catch above 99% of rewrites just before the sharpest boundary and 3.8% just after it. Detectors trained on later versions can also miss rewrites of earlier ones. Vocabulary differences between versions largely track where detection transfers and where it fails. In the two screening scenarios we simulated, screens covering all 23 versions either flagged one in eight human-written abstracts or missed one in three rewrites of the newest version. Indeed, a commercial detector missed most rewrites of the version just after the sharpest boundary while flagging almost no human-written abstracts. Research-integrity policy should therefore treat the benchmark accuracy of a detector as provisional, to be re-verified with every LLM release, including earlier versions.