Here is a sentence that should terrify every journalist quietly using ChatGPT to "polish" their copy: Alex Preston, a freelance book reviewer for The New York Times, just got fired because his AI tool lifted entire phrases from a Guardian review of the same book. Not paraphrased. Lifted.

Preston reviewed Jean-Baptiste Andrea's Watching Over Her for the NYT in January. A reader noticed the language felt familiar. Turns out it felt familiar because whole descriptions, including the phrase "lazy, Machiavellian Stefano" and a lengthy concluding assessment, were near-identical to Christobel Kent's August review of the same novel in The Guardian.

When confronted, Preston admitted he used an AI tool that "incorporated material from the Guardian review into his draft, which he failed to identify and remove." The Times appended an editor's note, cut ties with Preston, and called it "a clear violation of the Times's standards."

The Real Story Is Not About One Freelancer

Preston is embarrassed. He should be. But the bigger issue is that AI tools routinely regurgitate copyrighted material and present it as original output. This is not a bug. This is what these models do. They were trained on the internet, including Guardian book reviews, and they will happily feed that material back to you without a citation in sight.

The New York Times, ironically, is currently suing OpenAI for exactly this kind of content reproduction. And now their own freelancer got caught doing the same thing with a different AI tool. You cannot make this up.

Preston told The Guardian he was "hugely embarrassed" and called it "a serious mistake." Fair enough. But the mistake was not using AI. The mistake was trusting AI to produce original work and then not checking. That is the mistake thousands of writers, students, and professionals are making every single day. Preston just happened to get caught.

What Happens Next

Every newsroom is about to have a very uncomfortable conversation about AI policies. Some already have. The NYT explicitly bans AI-generated content. But enforcement depends on the honour system, and honour systems collapse the moment the tool is faster than the human.

This will not be the last time an AI plagiarism scandal hits a major publication. It will not even be the last time this month. The tools are too easy to use, the output too convincing, and the pressure to produce too relentless. The only question is whether publications will build real detection systems or keep pretending their freelancers are not using the same tools everyone else is.

First reported by The Wrap. Confirmed by The Guardian and The New York Times.