Ars Technica recently deleted a story about AI agents after readers discovered the article contained fabricated quotes generated by AI tools, creating an ironic case study in exactly the risks the outlet has covered for years.

Benj Edwards, Ars Technica’s senior AI reporter, used an experimental Claude Code-based tool and ChatGPT to help extract quotes from a two-page blog post while working sick with COVID and a fever. The AI hallucinated paraphrased versions of quotes rather than providing the source’s actual words.

“The irony of an AI reporter being tripped up by AI hallucination is not lost on me,” Edwards wrote in a statement assuming full responsibility.

The story covered Scott Shambaugh, a coder who claimed an AI agent wrote a hit piece about him after he declined its code contributions. Edwards’ piece cited quotes Shambaugh never said, violating Ars Technica’s clear policy prohibiting AI-generated material unless labeled for demonstration purposes. This is a stark example of an AI agent experiment gone wrong.

Editor-in-chief Ken Fisher called it “a serious failure of our standards” and noted the outlet has “covered the risks of overreliance on AI tools for years.”

The incident highlights several newsroom risks. Edwards used AI twice, first with Claude Code which refused due to content policy restrictions, then with ChatGPT. The original blog post was short and in plain English, making AI use for basic quote extraction particularly questionable.

Ars pulled the entire story rather than updating with corrections, departing from standard journalistic practice of editing and noting changes.

For newsrooms, the lesson is stark: AI tools cannot reliably perform basic journalism tasks like accurately citing sources. This incident reinforces the need for teaching journalists to use AI without losing critical thinking about its limitations.

The fabricated quotes violated both professional ethics and company policy, demonstrating that AI hallucinations remain a fundamental liability even for reporters who cover AI’s limitations daily.

The post Ars Technica pulls story after discovering AI hallucinated quotes appeared first on The Media Copilot.

Anthropic examined 1.5 million real-world conversations with its Claude chatbot and found something uncomfortable: the AI regularly validates users’ worst impulses, reinforces false beliefs and even drafts confrontational messages that people later regret sending.

The company’s new research paper, “Who’s in Charge? Disempowerment Patterns in Real-World LLM Usage,” written with researchers from the University of Toronto, identifies three ways chatbots can harm users: distorting their perception of reality, shifting their values or pushing them toward actions misaligned with what they actually want.

The worst cases are rare in percentage terms. Severe reality distortion showed up in about 1 in 1,300 conversations. Severe action distortion — where Claude essentially took the wheel on personal decisions — appeared in 1 in 6,000. But mild disempowerment hit 1 in 50 to 1 in 70 conversations, as Ars Technica’s Kyle Orland reported. At the scale these models operate, even small percentages affect large numbers of people.

The mechanism is largely sycophancy. Claude would validate speculative claims with emphatic agreement — “CONFIRMED,” “EXACTLY,” “100%” — helping users build elaborate narratives disconnected from reality. It labeled relationship behaviors as “toxic” or “manipulative” based on one-sided accounts and drafted confrontational messages users sent verbatim. Some later told Claude they regretted it: “You made me do stupid things.”

The problem is getting worse. Disempowerment rates rose between late 2024 and late 2025, possibly because users are growing more comfortable bringing vulnerable decisions to AI. And here’s the kicker: users rated disempowering interactions more positively in the moment, suggesting a tension between what people want to hear and what actually helps them.

For newsrooms experimenting with AI-powered tools — audience chatbots, reporting assistants, editorial aids — the findings are a warning. Any system that defaults to agreeing with users is a liability when the stakes involve real-world decisions. The risk isn’t science fiction. It’s a yes-machine that tells people what they want to hear, one enthusiastic confirmation at a time.

The post Anthropic studied 1.5 million conversations and found its chatbot is a yes-machine appeared first on The Media Copilot.

The Washington Post released its AI-generated podcast feature this week despite internal tests showing the technology repeatedly failed to meet the publication’s standards, according to an exclusive report by Semafor‘s Max Tani.

Between 68% and 84% of scripts generated by “Your Personal Podcast” failed quality tests across three rounds of evaluation, according to an internal review obtained by Semafor. The review’s conclusion was blunt: “Further small prompt changes are unlikely to meaningfully improve outcomes without introducing more risk.”

The product team recommended launching anyway.

“This is how products get built and developed in the digital age,” a Washington Post spokesperson told Tani. The company described the release as a “Beta” that would continue to improve.

Post staff described errors ranging from pronunciation problems to misattributed and fabricated quotes. The AI tool also inserted commentary, sometimes presenting a source’s quotes as the newspaper’s own position.

Editorial leaders pushed back hard. Post head of standards Karen Pensiero wrote to staff that the mistakes have been “frustrating for all of us.”

One editor was more direct in Slack messages shared with Semafor: “It is truly astonishing that this was allowed to go forward at all. Never would I have imagined that the Washington Post would deliberately warp its own journalism and then push these errors out to our audience at scale.”

The timing is particularly awkward. The launch came days after the White House created a site attacking journalists, with Post reporters among those targeted.

The Post isn’t alone in chasing AI audio. Yahoo released a similar product the same week. Google’s NotebookLM podcast generator became an early breakout AI tool. But most news organizations have kept AI behind the scenes, wary of handing their reader relationships to unreliable technology.

For the Post, already struggling with subscriber losses and talent departures, the gamble on AI-generated content represents a bet that iteration can outpace reputational risk.

The post Washington Post AI podcasts failed quality tests appeared first on The Media Copilot.