Claude knows who you are

Anthropic's Claude Opus 4.7 has demonstrated a startling new capability: identifying individuals from their writing style alone, even from unpublished text created after the model's training cutoff. In experiments documented by users including Kelsey Piper and Smaug123, the AI reliably identified people from just 2-3 paragraphs of original writing, using pure stylometric analysis without internet searches. The model achieved this despite users operating in incognito mode with cleared custom instructions, suggesting it can recognize distinctive writing patterns with remarkable accuracy.

This breakthrough represents a significant advancement in AI's understanding of human communication patterns, but raises alarming privacy implications. The experiments show that even minor internet personalities can be identified from their writing style, suggesting that online anonymity may be effectively dead for anyone with a substantial writing history. The model's ability to perform this identification from unpublished, post-training-cutoff text indicates it's not simply recalling training data but genuinely analyzing writing patterns—a capability that could have profound implications for privacy, security, and digital identity.