Is This What Comes After AI Slop?

Summary

H. M. Wolfe's viral novel 'Daggermouth,' themed around a surveillance state, may be the first bestseller largely written with AI assistance, according to a Stony Brook University study. The research found that 60% of the text was AI-generated and revealed that it shares exact phrases with other suspicious AI e-books.

Why it matters

This incident forces publishers, readers, and detection researchers to confront the question of whether 'AI slop' can produce genuinely compelling fiction that outperforms human writing. As AI-assisted books constitute 20% of Amazon's e-book catalog and 10% of genre-specific bestsellers, the human-machine authorship boundary is becoming increasingly significant commercially and culturally. This situation also challenges tools like Pangram, whose acknowledged false positive rates make definitive attribution nearly impossible.

Highlights

• A pattern described by Chakrabarty as 'too many telltale signs' repeats throughout 'Daggermouth.' This pattern consists of five-word 'rare phrases' that are identical to those found in other suspicious AI e-books.
• Mohit Iyyer from UMD states that a 60% AI detection rate for a fully human-written book is 'almost statistically impossible,' even when accounting for Pangram's margin of error.
• Unlike mass-producing scammers (such as an author with 418 books), only two novels are registered under Wolfe's name. This suggests that genuine human curation and editing were layered on top of the AI draft.

Critical take

The study has not yet undergone peer review. Pangram's false positive rate is explicitly acknowledged. Wolfe and Simon & Schuster did not respond to the evidence of identical phrases. Therefore, despite journalistic diligence, the report presents strong circumstantial evidence rather than definitive proof. Consequently, the claim of being the 'first AI bestseller' cannot go beyond informed suspicion.