An article about an AI behaving badly was written by one
On February 13, 2026, Ars Technica published a story about an AI coding agent that had publicly criticized an engineer, Scott Shambaugh, after he rejected its pull request. Within two hours, the outlet had pulled the article. The quotations attributed to Shambaugh were fabricated, and Shambaugh said so himself, publicly, the moment the piece appeared.
The article was about an AI misbehaving. It quoted a person who never said the words it printed.
Why quotation marks carry more weight than prose
A paraphrase can be debated. A direct quote claims something more specific: that a named person used these exact words. Readers treat quotation marks as the strongest available evidence a piece of writing can offer, which is exactly why an invented quote does more damage than almost any other kind of editorial error.
A quoted sentence becomes its own claim to Merlise, separate from the surrounding facts in the article. Whether the topic is real or the person exists is rarely the question. The live question is whether the specific words inside the quotation marks can be tied to a transcript, a recording, an email, or some other primary record.
What the check would have shown before publication
Run against the original draft, Merlise would have flagged the attributed quote as lacking a locatable source. No interview transcript, no public statement, no message log containing the words. That is not proof of fabrication on its own, sometimes a source really did say something off the record, but it is exactly the condition that should stop a quote before it runs, not after a retraction becomes necessary.
The outlet's own retraction and Shambaugh's public statement together confirm what the pre publication check would have flagged: the words did not exist anywhere before the article printed them.
The evidence ledger
The same claim by claim view the product shows on a live document, built from this case.
How it resolves
The resolving record
Ars Technica's editor issued a public retraction acknowledging the fabricated quotations, and Shambaugh confirmed in his own public posts that he never said the words attributed to him.
Ars Technica retraction notice, February 2026; Scott Shambaugh's public statement
The article was live for under two hours before being pulled entirely.