Ars Technica has fired longtime AI reporter Benj Edwards after he published fabricated quotes generated by ChatGPT in a story, mistaking them for real statements scraped from an engineer's website.
The sequence of events is almost painfully ironic. Edwards was writing about an AI agent that had launched a smear campaign against Scott Shambaugh, a coding repository volunteer who had rejected AI-generated code. While reporting the piece, Edwards tried to use Claude to pull quotes from Shambaugh's personal website but was blocked by the site's code. He switched to ChatGPT, which returned quotes that looked plausible but were entirely made up. Edwards published them without checking whether Shambaugh had actually said any of it.
How a Routine Shortcut Turned Into a Firing
The fabricated quotes were eventually caught. Ars Technica issued a retraction and an editor's note apologizing to Shambaugh. Edwards publicly acknowledged the mistake, saying he had been sick with the flu while writing the piece. By February 28, he was out. His author bio on the site was quietly changed to past tense. No public explanation from the publication.
Edwards said he was "told by management not to comment until they did." As of this writing, Ars Technica hasn't made a public statement beyond the retraction.
The Real Problem Isn't One Reporter
This is a case study in what happens when AI tools get woven into journalistic workflows without guardrails. ChatGPT and similar large language models hallucinate - they generate text that reads like fact but is pulled from statistical patterns, not reality. Using them to "extract" quotes from a website is fundamentally misunderstanding what these tools do. They don't read and copy text. They predict what text should look like.
Edwards, to his credit, was one of the more knowledgeable AI reporters in tech media. He wasn't some newcomer who didn't understand the technology. That makes this worse, not better. If someone who covers AI daily can fall into this trap while rushing through a story, the failure mode is systemic, not personal.
Newsrooms are under pressure to produce more with fewer people. AI tools promise speed. But verification - actually checking whether a quote is real - is the one step you absolutely cannot automate with a model that makes things up by design. The lesson here isn't "don't use AI in journalism." It's that AI cannot replace the parts of journalism that require confirming what's true.