The New York Times and the Critic Who Borrowed a Sentence from a Machine

The New York Times has parted ways with critic Alex Preston following his admission that he used an artificial intelligence tool while writing a January review of Jean-Baptiste Andrea's Watching Over Her. The AI, it emerged, had helpfully contributed a phrase that bore an unfortunate resemblance to a Guardian review by Christobel Kent. Preston apologized. The Times declined to continue the relationship.

This is, on the surface, a tidy story: critic uses AI, critic gets caught, critic says sorry. The Times takes the high road. Everyone moves on. The mechanism of accountability works more or less as designed, which is reassuring, given how rarely it does.

But sit with the more interesting question. Preston is not, by any account, a lazy writer. The use of an AI tool to assist in the drafting of criticism is not, in principle, categorically different from using a thesaurus, consulting a summary when your copy of the book is buried under a pile of galleys, or asking a colleague to read a draft. The problem is not that a machine was involved. The problem is that the machine plagiarized someone else, and Preston either failed to notice or failed to check.

What the incident really exposes is the specific way AI tools fail in the one context where they seem most useful: the precise, high-accountability production of text. A machine can generate plausible-sounding prose about almost any book. What it cannot do is tell you whether that prose is original, borrowed, or assembled from patterns in its training data that happen to overlap with a Guardian review published three weeks earlier. This is not a malfunction. It is how these systems work.

Andrea's novel — which won the Prix Goncourt in 2023 under the French title Veiller sur elle — deserves better than to be the vehicle for someone else's lapse in judgment. So does Preston, perhaps, if one is in a generous mood.

The New York Times will now find someone else to cover the books Preston was assigned. The review section will continue. This episode will be referenced in essays about AI and journalism for the next several years. And somewhere, a critic with no interest in cutting corners will sit down with a novel they have actually read and write something nobody else wrote first. That seems like the right outcome. Whether it is also a scalable one is another question entirely.