Show Me How You Reason and I'll Tell You Who You Are: Reasoning Graphs for Robust LLM Authorship Attribution

Researchers have developed a method for authorship attribution that analyzes the reasoning patterns within large language models rather than just surface-level linguistic features. By creating reasoning graphs that map how models reach conclusions, this approach aims to identify the specific LLM behind a piece of text more reliably. This technique offers a way to distinguish between different models as their generated outputs become increasingly similar in style and structure.