Report: Writers, academics say AI tools are flagging their decades-old work as machine-generated

A viral thread on X has writers and academics sharing their discovery of AI detection tools flagging their original work — some written long before AI even existed — as machine-generated, fueling speculation that their writing was quietly scraped and used to train AI models without their consent.

The thread took off after British author Adam Kay posted on March 21 that when he ran his writing through an AI checker, it flagged 29.7% of the text as AI-generated. 

"Thing is, it obviously wasn't," Kay wrote. "Ethical and creative reasons aside, the book in question is nearly a decade old — well before technology could do this."

Kay, best known for the memoir This Is Going to Hurt — a dark comic account of his years as a junior doctor that was adapted into a British Academy Film and TV Award-winning series — was not alone.

Paul Spicker, a British social policy professor, posted that he tested his first academic article, published 45 years ago. It came back 77% AI-generated.

Epidemiologist Devi Sridhar, a professor and chair of global public health at the University of Edinburgh, went further in a March 21 X post, directly naming Anthropic, the company behind the Claude AI chatbot.

"I know my previous books and articles have been used to train AI (looking at you anthropic)," Sridhar wrote, adding that her pre-AI articles now come back "as high as 90% AI" in detection tools.

 "It's not artificial intelligence - it's collective human intelligence," she said.

Also joining the thread was political commentator and author Gabrielle Perry, who has bylines in the Washington Post and Essence. She noted March 21 that her publisher, Hachette, warned her not to   write her book in Google Docs for this reason.

“Google is training its AI on everything you type in it," she wrote, "which makes these things hit or miss."

One commentator underscored the absurdity of the situation by posting a screenshot of an AI detection tool flagging president Abraham Lincoln's Gettysburg Address as AI-generated.

Similarly, another X user posted the opening of chapter 5 from Mary Shelley’s Frankenstein flagged as a “100% AI generated” passage. 

Future litigation concerns 

The concern about academic work being scooped up to train AI is not unfounded. According to Nature, virtually all academic researchers should assume their published papers have been used to train AI models. The magazine reported in 2024 that millions of academic papers scraped from the internet appear in the datasets used to build large language models, and that publishers have sold access to research to tech firms — often without telling the authors.

The U.S. Copyright Office released a 108-page report in May 2025 concluding that some uses of copyrighted material to train AI cannot be defended as fair use — a ruling that lends weight to lawsuits several authors have already filed against AI companies.

Meanwhile, AI detection tools themselves remain notoriously unreliable. Research has found that human-written scientific manuscripts are routinely misidentified as AI-generated, with false positive rates particularly high for non-native English speakers and writers with formal, structured styles — the very qualities that define academic prose.

Those false positives carry real consequences. Universities and journals are increasingly leaning on tools like Turnitin's AI detector to police submissions — tools that, by researchers' own assessments, cannot be trusted to make that call accurately.