Researchers debate reliability of tools spotting AI-written novels after bestseller dispute

In an interview, SUNY Stony Brook computer scientist Tuhin Chakrabarty described his work on detecting AI-generated text and highlighted a high-profile dispute over the 2026 bestseller Daggermouth, which a journalist alleged relied heavily on AI despite the author H. M. Wolfe’s denial. Chakrabarty’s preprint examined rare word strings that appear verbatim in large corpora, analyzing 200 AI-flagged titles, 200 self-published bestsellers, and 200 award-winning literary novels.

The analysis showed AI-written books shared the greatest overlap with existing literature, reflecting the ease with which models reproduce formulaic genre conventions. He noted that longer, highly specific phrases are unlikely to arise by chance, making them useful detection markers. While acknowledging that human writers can also repeat common expressions, Chakrabarty argued that undisclosed AI use raises moral and copyright concerns. He also discussed legitimate creative uses of AI, citing examples such as Vauhini Vara’s nonfiction work and Sheila Heti’s New Yorker story, and suggested future markets may differentiate pure human, hybrid, and fully AI-generated writing.