Dan Bernstein of Sportico posted on X on May 17 to accuse Sports Illustrated of “stealing entire stories from people without credit, seemingly using AI.” Within 90 minutes the offending article had been deleted, the byline of a reporter named Parker Loverich had disappeared from SI.com, and Loverich’s LinkedIn and X accounts had gone dark. The piece in question, published on May 15, recycled the data and analysis from a Sportico article Bernstein and Lev Akabas had published two days earlier on the 100 million US dollars American gamblers had lost on Kalshi parlays so far this year. SI later told Awful Announcing that the prediction-market vertical was run by “an independent publisher” whose ties had now been cut. The magazine had been in this position before, in November 2023 under different ownership, when it allegedly published articles by fake reviewers with AI-generated headshots. The reputational damage from that earlier episode took perhaps six months to fade.
According to Brittany Allen, a writer and actor, Ethan Mollick, a Wharton professor who studies the workplace effects of generative AI, ran Jamir Nazir’s prize-winning short story, The Serpent in the Grove, through Pangram, widely regarded as the most reliable AI-generated content detector, “on a hunch” in mid-May. The verdict came back as machine-generated. A judging panel chaired by the novelist Louise Doughty had selected the story as the Caribbean regional winner of the 2026 Commonwealth Short Story Prize from 7,806 entries. Granta, which hosts the prize winners on its website in partnership with the Commonwealth Foundation, posted a statement saying it would keep the story online “until definite evidence comes to light”.
In an essay for the Atlantic, Vauhini Vara, a Pulitzer-shortlisted novelist, noted that before long, commenters were “pointing fingers” at two other winners of this year’s Commonwealth Prize: Malta’s John Edward DeMicoli and India’s Sharon Aruparayil. The writers have, by and large, said nothing. It marks a shift. Writers named in earlier episodes by the New York Times, Sports Illustrated and Hachette typically issued statements as a matter of reputational survival.
Hachette withdrew the novel Shy Girl by Mia Ballard in April after the New York Times brought passages and detector readings to its attention. Pangram had classified a substantial part of the manuscript as AI-generated. Ballard told the Times she had not used AI herself, and that an editor she had hired during the self-published phase of the book may have inserted machine-written passages without her knowledge. Hachette cited its commitment to “original creative expression and storytelling”. A three-hour video by the YouTube channel Frankie’s Shelf dissecting the novel had been posted in January 2026 and has so far been viewed more than 1.4 million times. The publisher acted only after the Times got involved.
Reputational discount
Peter Vandermeersch, a senior journalist who had previously edited the Dutch newspaper NRC, was suspended by the publisher Mediahuis in March after NRC reported that he had inserted dozens of fabricated, AI-generated expert quotes into 15 of 53 articles filed for two Mediahuis sites. Coverage outside the Netherlands lasted under two weeks. The affected articles, where identifiable, were corrected and left online rather than withdrawn.
Vandermeersch wrote on Substack that he had relied on tools including ChatGPT, Perplexity, and Google’s Notebook to summarise lengthy reports, trusting the outputs to be accurate. Instead, the systems had invented quotes that ultimately “put words into people’s mouths,” he said.
Nathan Wolfrath and six co-authors published a longitudinal analysis in JAMA Network Open in March of 7,251 articles in the journal from January 2022 to March 2025. The monthly proportion of papers in which the body text was flagged as AI-generated rose from zero to 11.3 per cent over the period. Only 0.2 per cent of authors had disclosed any large-language-model use, and 40 per cent of those who did disclose had then been flagged by the detector. The authors stressed that detector accuracy is imperfect and that the figures should not be read as definitive. The pattern is still notable. In a journal that explicitly asks authors to disclose AI use, the share of flagged articles has risen sharply, and disclosure has not kept pace.
YouGov and Meltwater surveyed nearly 10,000 consumers across seven markets in April 2026 and found that most remained sceptical of AI-generated content, and that the scepticism affected brand trust. The Nuremberg Institute for Market Decisions ran a separate test on 1,000 respondents apiece in America, the UK, and Germany, presenting identical material with and without an AI-generated label. Readers told the content was machine-written engaged with it less, but subscription cancellations and traffic drops on the scale that would discipline publishers have not, on current evidence, materialised.
Steve Cannella, the editor-in-chief of Sports Illustrated, told Front Office Sports on May 21 that the company runs around 200 affiliate sites on the SI.com domain. The sites benefit from the parent brand’s search ranking and operate under separate editorial control. The cost of an occasional cancelled article and a deleted byline is modest. The upside is the volume of search-driven traffic the affiliate network produces. Granta, similarly, did not commission Nazir’s story; the Commonwealth Foundation chose it, and Granta hosts it.
Seven co-authors at the University of Maryland and Pangram posted a preprint to arXiv in October 2025 that scanned 186,000 articles published by 1,500 American newspapers in the summer of that year. Pangram flagged 9.1 per cent of them as partially or fully AI-generated. Manual review of 100 flagged articles found that five had disclosed AI use. Seven of the 1,500 newspapers had a published AI policy. Detection now works at scale, as the study demonstrates. Disclosure has not kept pace, and readers have not yet made publishers change course.
Photo: Dreamstime.
