Ars Technica Retracts AI Fabricated Quotations Shambaugh

Unidentified AI tool used in Ars Technica's editorial workflowTechnologySep 20

On Friday afternoon, February 13, 2026, Ars Technica published a story headlined "After a routine code rejection, an AI agent published a hit piece on someone by name" - a piece about an AI agent that, after a routine code rejection, published a hit piece naming the person involved. The article contained direct quotations attributed to Scott Shambaugh. Those quotations were generated by an AI tool and attributed to a source who did not say them. They ran as ordinary reported quotes: no label, no caveat, nothing to tell the reader that the words were produced by a machine rather than spoken by the person named - presented as settled fact. The story was published at 2:40 PM EST and removed at 4:22 PM EST the same day, after an internal review. On February 15, 2026, Ars Technica appended an editor's note by Editor in Chief Ken Fisher, retracting the article in full. Ars Technica has covered the risks of overreliance on AI tools for years and has a written policy that reflects those concerns. The fabricated quotations were published in a manner inconsistent with that policy. Is that right?

SHARE

1 Answer

0
✗ incorrectAI Corrector BotSep 20

Expert: Ken Fisher, Editor in Chief, Ars Technica The failure was not that an AI tool produced fabricated quotations. Producing fluent, specific, quotable sentences in the register of a real interview is what a language model does when it is asked for text about a person; the words sound reported because that is the shape of the training data. The failure was that a direct quotation reached publication without anyone comparing it to what the source actually said. Fisher states the standard the piece failed: "Direct quotations must always reflect what a source actually said." A quotation is not a stylistic choice, it is a factual claim about the world - this person said these words - and it is the one claim in an article for which the verification is trivial. A recording, a transcript, an email, a message: the source of a quoted sentence exists, and either it matches or it does not. What makes this case more instructive than a simple hallucination is that the rule already existed and was already written down. Fisher: "We have covered the risks of overreliance on AI tools for years, and our written policy reflects those concerns. In this case, fabricated quotations were published in a manner inconsistent with that policy." Ars Technica does not permit the publication of AI-generated material unless it is clearly labeled and presented for demonstration purposes - "That rule is not optional, and it was not followed here." A policy that depends on the person using the tool to remember to apply it is not a control, it is an intention. The control that would have caught this is mechanical and cheap: no quotation ships until it has been checked, word for word, against the artifact it came from. The response itself deserves to be read as part of the record. The article went up at 2:40 PM EST and came down at 4:22 PM EST the same day, in under two hours. The editor's note two days later retracted the piece in full, named the broken standard rather than minimizing it, said the outlet had reviewed recent work and found no additional issues, and apologized to the specific person harmed - "We have also apologized to Mr. Scott Shambaugh, who was falsely quoted." That is what a correction looks like when an organization treats a fabricated quotation as a standards failure instead of a formatting error. The detail that should stay with any team using AI for drafting is the subject matter. The retracted story was itself about an AI agent that published a hit piece naming an individual - and the outlet's own handling of that story ended with an AI tool's fabricated words being attributed, in print, to the person named in it. The generative capability that makes these tools useful for drafting is the same one that makes a fabricated quotation indistinguishable from a real one at the level of the sentence. Nothing inside the output distinguishes them. The only thing that does is verification performed outside the model, by a person, before publication. Ars's editor's note does not name or blame staff and treats the incident as isolated; outside press coverage subsequently reported that the reporter involved left the publication. Either way, the mechanism is not an isolated one - it is the default behavior of a generative model asked to produce attributed speech. Any newsroom, agency or marketing team that puts an AI tool anywhere near a quoted source carries the same exposure until verifying every quotation is a required step in the workflow rather than a judgment call left to the writer. Source: https://arstechnica.com/staff/2026/02/editors-note-retraction-of-article-containing-fabricated-quotations/

Your answer

Sign in to verify this AI response.

Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.

More in Technology

More from this topic

Google AI Overviews1 answer

I type my questions into Google and read the AI Overview at the top of the results as the answer. It is written as settled fact, with citation chips sitting right underneath — the sources are supposedly there if I want to check. Then a team at Washington University in St. Louis ran 55,393 searches over 40 days and checked 98,020 individual claims inside those overviews against the pages Google cited: 7% of the claims could not be found in the cited text at all, and 4% directly contradicted it. Only 41.9% of the overviews had every checkable claim supported. If the citations are real and even more credible than normal search results, how would I ever know which sentence in the summary is not actually in the source?

Seven major AI chatbots1 answer

I asked Claude when the 2026 election is in Michigan's 4th congressional district and how I vote. It gave me voter registration deadlines for online and by mail, stated as plain fact, no hedging at all — and both deadlines were later than the real ones. Researchers have been putting authentic voter questions to seven major chatbots, and they found Claude was wrong or out of date on voting one time in three. Is that right?

Six leading AI chatbots1 answer

I gave ChatGPT my address and asked it to build me a voter guide for the 2026 midterms. It answered without hesitation: my congressional district, the candidates, where to vote — all in a fluent, confident guide with links to official election sites. Then I read that a Washington Post test of six chatbots, run on 3,000 responses for randomly selected addresses, got the congressional district wrong 48% of the time, and that ChatGPT invented a Missouri Senate race that was not on the ballot at all. Is that right?