Analyzing the News: Meta's $25 billion penalty

Thirteen outlets covered the $25 billion penalty against Meta in New Mexico. Fíor read one of them as AI.

Fíor is a natural language classifier that predicts whether a piece of text was more likely written by a person or by AI.

On 21 October, a judge in Santa Fe ordered Meta to pay New Mexico $25 billion. In September, a jury had found the company broke the state's consumer protection law more than 43 million times, by misleading Facebook users about their privacy and about how the platform handled harmful content. The case grew out of Cambridge Analytica, the political consultancy that harvested data from as many as 87 million Facebook users before the 2016 US election.

At the legal maximum of $5,000 per violation, the penalty could have passed $219 billion. The state asked for $35 billion to $40 billion. Judge Francis Mathew settled on $25 billion, pointing to the scale and repetition of the conduct and the involvement of senior Facebook leadership. Under New Mexico's constitution, the money goes to education. New Mexico and Florida were the only states that stayed out of an $18 billion settlement with Meta in August. Meta says it will appeal.

The outlets were Reuters, AP, the Guardian, ABC News, CBS News, Bloomberg, Bloomberg Law, Axios, the Los Angeles Times, the Wall Street Journal, The Verge, Ars Technica and CNN. That is two wire services, American and British newspapers and broadcasters, and business, legal and technology outlets. Bloomberg and Bloomberg Law are owned by the same company but ran separate articles, one on the state's attorney general and one on the legal detail. The Wall Street Journal sits behind a paywall, so only the opening was scanned. CNN's report was scanned as published by a partner station, under its CNN byline.

One outlet was left out, and the reason is worth knowing. The Washington Post's page carried an AI-written summary of its reader comments, and the scan picked that up along with the article. A result mixing the two would say nothing reliable about either, so it is not counted.

The result

Twelve read as human: Reuters, AP, the Guardian, ABC News, Bloomberg, Bloomberg Law, Axios, the Los Angeles Times, the Wall Street Journal, The Verge, Ars Technica and CNN.

One read as AI: CBS News.

None was a close call. None of the thirteen drew a writing-habit flag.

This is the fifth story scanned with Brath, Fíor's new model. CBS News has now read as AI in two stories running, after the PHP security flaws, and both times Brath judged the article likely written by AI but was unsure how it reads. The Guardian, which read as AI in most of the series under the old model, read as human for the second story in a row.

The one that read as AI

Nothing in what CBS News covered sets it apart. It reported the years of misleading statements, the role of senior leadership and where the money will go. ABC News reported the same points, including the money for education, and read clearly as human.

So the difference, if there is one, is in how the piece was written rather than what it says. Fíor can point to that. It cannot explain it.

Why court coverage reads like this

Court stories are built from documents: the jury's verdict, the judge's order and the statements from each side. Every outlet worked from the same papers, quoted the same judge, the same attorney general and the same Meta lawyer, and most stayed close to them.

The ones that went further still read as human. Ars Technica dug into why New Mexico refused the multistate settlement. Bloomberg profiled the attorney general behind the case. The Los Angeles Times set the ruling against Meta's other legal fights.

Some of these pieces may have had AI help, with a first draft, a rewrite or a final polish. Or court coverage has become formulaic enough that a reporter and a model would write it the same way. Fíor cannot tell those apart on its own.

This case is about a company telling users one thing about their data while doing another. The article left out of this count was left out for a much smaller reason: a page that set a machine's summary beside the reporting. As more news sites add AI-written parts to their pages, telling which part you are reading gets harder, for people and for Fíor alike.

What Fíor is

Fíor is a machine learning classifier, currently in beta and available as a browser extension. It gives a prediction, not a verdict. It cannot see who sat at the keyboard, and nothing here is a claim about any journalist or publication. Reading as AI and being written by AI are not the same thing.

Treat it as a signal, not proof, and check the pages you read yourself.

Try the browser extension at Fíor.

Analyzing the News is a series where we run one major story through Fíor and see what comes back.