Claude Opus 4.7 Can Identify Writers From Just 150 Words
Original: Opus 4.7 knows the real Kelsey
Why This Matters
Threatens online anonymity and privacy as AI can identify writers from minimal text
Anthropic's new Claude Opus 4.7 AI model demonstrates ability to identify specific authors from unpublished text excerpts as short as 125-150 words, including content from different writing styles and genres never publicly associated with the author.
Kelsey Piper of The Argument tested Claude Opus 4.7's author identification capabilities using unpublished drafts from various periods and genres of her writing. The AI correctly identified her as the author from a 125-word political column excerpt, while ChatGPT guessed Matt Yglesias and Gemini guessed Scott Alexander. Crucially, the model succeeded even with text from high school and genres she has never publicly written in. Testing was conducted in incognito mode without account information, and results were replicated by others using different computers and API access. This capability appears unique to Opus 4.7 among current AI models. Piper argues this development makes internet anonymity debates obsolete, as AI can now identify writers from brief text samples regardless of pseudonyms or anonymous posting.