Proving Claude Knows You by Your Code Style: Author2Vec
Author2Vec is a tool that analyzes coding style to identify the author of source code. By examining stylistic features like naming conventions, formatting, and syntax patterns, it can attribute code to specific developers, demonstrating that programming style is as unique as a fingerprint.
Background
- Author2Vec is a technique for identifying the author of a piece of code — similar to how stylometry can identify writers from their prose, but applied to programming style (variable naming, spacing, comment patterns, etc.).
- The demo page suggests that Claude, an AI assistant built by Anthropic, is being tested to see if it can recognize a user's coding style from past conversations and adapt its responses accordingly.
- This matters because it points toward AI systems that build persistent "profiles" of individual users, raising questions about personalization vs. privacy, and whether users are aware that their stylistic fingerprints are being recorded and used.
- The project draws on earlier research in code authorship attribution (like Chen et al.'s "Code Stylometry") and the broader trend of AI assistants that remember user habits across sessions.