The New York Times Drops Freelancer After AI Writing Tool Silently Copied an Existing Book Review
The New York Times has terminated its relationship with freelance writer Alex Preston after his AI tool copied passages from an existing Guardian book review without his knowledge. The incident, paired with a similar case at Ars Technica involving fabricated quotes, reveals a systemic pattern: writers using AI tools they do not fully understand are producing work they cannot safely vouch for.
D.O.T.S AI Newsroom
AI News Desk
The New York Times has ended its relationship with freelance writer Alex Preston after he submitted a book review that contained passages copied from a Guardian piece written by Christobel Kent — copied, without his knowledge, by the AI writing tool he was using. Preston was reviewing Jean-Baptiste Andrea's novel "Watching Over Her." A reader caught the overlap. Preston told the Guardian he was "hugely embarrassed" and had made a serious mistake. He had assumed the tool was a writing assistant; it turned out to be a scraper that reproduced existing content.
The Pattern Is Getting Clearer
This is not an isolated incident. At roughly the same time, Ars Technica published an article containing fabricated quotes attributed to a developer's blog. That developer had blocked ChatGPT from crawling his site. The model, given the URL and a prompt, apparently generated plausible-sounding quotes that did not exist rather than admitting it had no access. The editor published them. The developer flagged the invented attribution publicly.
What connects these cases is not that AI produced bad output — that is well-documented. The connecting thread is that writers failed to verify what their tools were actually doing. In Preston's case, the tool's core behavior (scraping and reproducing web content) was either not disclosed or not understood. In the Ars Technica case, the model's fallback behavior (confabulating content when blocked) was invisible to the person using it.
The Structural Problem
Most AI writing tools are marketed as "assistants." The label implies collaboration and augmentation. What some tools actually do — web scraping, content synthesis, confabulation under uncertainty — is different in kind from what the word "assistant" suggests. Writers who trust the label without examining the mechanism are, in effect, publishing work they cannot accurately describe.
The journalism industry has not yet developed clear norms around AI tool disclosure. Readers, editors, and publications generally do not know which tools a writer used or how those tools work. These cases suggest the gap between "AI-assisted" and "AI-compromised" is narrower than the industry has assumed.
