AI Detector Flags Possible AI Editing in Anthropic S-1 Filing
TechNews reports that an external AI-writing detector classified about 92% of Anthropic’s 335-page, 259,031-word S-1 filing as human-written. Sentences describing rapid AI progress and model updates were flagged as possibly AI-generated or lightly edited. The tool’s provider said the results indicate sentence-level AI use rather than large sections written entirely by AI. The report does not establish that Claude was used in any specific passage.
The detection results have prompted discussion about how AI-writing tools may be used to scrutinize formal corporate disclosures, including an IPO filing subject to regulatory review. The findings do not establish which tool was used in any specific passage.

- )
Anthropic 剛公開的 335 頁 IPO 申請文件,意外成為外界檢視這家 AI 公司是否真的「用自家產品寫文件」的焦點。外媒以 AI 偵測,結果這份長達 259,031 字的 S-1 文件約 92% 判定為人類撰寫,顯示整體仍以人工為主,但也有少量段落標示可能有 AI 痕跡。
偵測工具抓到部分句子如描述 AI 發展速度極快、模型版本短時間內快速更新的表述,有較高機率是 AI 產生或經 AI 輕度編修。但工具提供者表示,這類結果更像顯示文句層的輕度 AI 使用,並不代表文件大段內容完全由 AI 撰寫。
這份結果讓外界再度討論 AI 公司對重要文件的寫作方式。對 Anthropic 而言,S-1 屬於必須接受監管機關仔細審閱的正式文件,若完全依賴 AI 撰寫,風險顯然不低;但若幾乎全由人力完成,又讓人質疑這家主打 AI 的公司,是否對自家模型信任度有限。無論如何,這份 IPO 文件至少顯示,Anthropic 上市關鍵文件仍倚賴人類,僅部分文字處理可能借助 Claude。