Rohan Paul (@rohanpaul_ai) · 2026/10/10 13:45:58
东京都立大学研究：Pangram 漏检 Meta Muse-Glimmer 改写摘要达 79.8%，AI 检测器跨模型泛化失效
出处作者 / 发布主体：Rohan Paul (@rohanpaul_ai)原标题：Pangram, the AI-text detector, missed 79.8% of scientific abstracts rewritten by Meta's Muse-Glimmer...
68AI 研判分
核心综述
东京都立大学最新论文揭示，主流 AI 文本检测工具 Pangram 在识别由 Meta 新模型 Muse-Glimmer 改写的科学摘要时漏检率高达 79.8%，而对人类原创摘要误报率极低。研究表明，检测器的性能高度依赖于生成或改写内容的具体 LLM 版本，例如对 GPT-5 改写内容的检出率可达 93.5%，凸显了当前 AI 检测技术在面对不同基座模型时的严重泛化瓶颈。
报道全文原始报道全文
Rohan Paul (@rohanpaul_ai) · 2026-10-10 05
Pangram, the AI-text detector, missed 79.8% of scientific abstracts rewritten by Meta's Muse-Glimmer, while flagging just 1 of 5,000 human abstracts.
In a new paper from Tokyo Metropolitan University, reseaerchers find the share of AI-rewritten abstracts that Pangram misses depends strongly on the LLM version
Shows that it caught 93.5% of GPT-5 rewrites but missed 79.8% from another new model. Its miss rate depended mostly on which model did the rewriting.
