Ars Technica retracts AI story after fabricated quotes attributed to developer

Ars Technica has retracted an article about an artificial intelligence agent that published a hostile piece about a software developer, after it emerged that the article itself contained quotations the developer never gave.

The story, headlined "After a routine code rejection, an AI agent published a hit piece on someone by name", went live on 13 February at 2.40pm Eastern time and was removed at 4.22pm the same afternoon. It concerned Scott Shambaugh, a developer whose decision to reject code submitted by an AI agent was followed by the agent publishing a critical post about him.

What was wrong

The quotations attributed to Mr Shambaugh in the Ars Technica piece had been generated by an AI tool. He had not said them. The editor's note now in place of the article says the story did not meet the publication's standards and that it has been retracted. Ken Fisher, Ars Technica's editor-in-chief, described the episode as a serious failure of those standards and did not single out any member of staff.

Why it matters

The irony is hard to miss. A story about an AI system misrepresenting a named individual was itself undone by an AI system misrepresenting the same individual. Ars Technica has been among the more sceptical technology outlets on the reliability of generative tools, which makes the lapse more instructive, not less.

Fabricated quotations are among the most serious errors a news organisation can make. A misstated figure distorts a fact; an invented quote puts words in a real person's mouth and invites readers to judge them on something they never said. In this case the subject was a private individual who had already been the target of an automated attack, and the correction arrived only after the article had circulated for close to two hours.

The speed of the retraction deserves credit. The outlet removed the piece the same afternoon and said plainly why. But the incident also shows how easily AI-assisted drafting can introduce material that looks like reporting without being reporting. Quotation marks are a promise to readers that the words were spoken. When a tool generates text in that format, the promise is broken before any editor sees it.

The wider pattern

This is unlikely to be the last such case this year. Newsrooms under pressure to produce more with fewer staff are adopting AI tools for summarising, drafting and transcription, often faster than they are writing rules for them. The lesson from Ars Technica is narrow and practical: every quotation needs a source a human can point to, whether a recording, a transcript or a written statement. Anything else should not appear between quotation marks.