Dissecting AI-related Paper Retraction Across Countries and Institutions

Research integrity is currently besieged by a surge in synthetic manuscripts. A forensic workflow is operationalized herein to isolate and quantify ``computer-aided'' misconduct within the global scholarly record. A corpus of \( N=3,974 \) retracted DOIs sourced from the Retraction Watch Database was analyzed, with records cross-linked to institutional metadata via the OpenAlex API. Through the application of fractional attribution modeling and the calculation of Shannon entropy (\( H \)) for retraction rationales, a distinct geographic schism in fraud typologies was identified. High-output hubs, specifically China and India, exhibit high reason entropy (\( H > 4.2 \)), where ``Computer-Aided Content'' frequently clusters with established ``Paper Mill'' signatures. These AI-driven retractions exhibit a compressed median Time-to-Retraction (TTR) of \( \sim \)600 days, nearly twice as fast as the \( 1,300 \)+ day latencies observed in the US and Japan---where retractions remain skewed toward complex image and data manipulation. The data suggests that while traditional fraud has not been replaced by generative AI, it has been effectively industrialized. It is concluded that current post-publication filters fail to keep pace with the near-zero marginal cost of synthetic content, necessitating a shift toward provenance-based verification.