Analyzing the News: The Google Cloud outage

Thirteen outlets covered the Google Cloud outage that hit British banks on Tuesday. Fíor read all thirteen as human.

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

On 20 October, what Google called a power event took one zone of its London cloud region offline. Problems started just after 7am UK time, and Google declared the incident over by 3pm, though it said a very small number of customers were still affected. In between, customers of Revolut, Lloyds, HSBC and Halifax had trouble with apps, card payments and withdrawals. Royal Mail, DPD, Rightmove and the courts' video platform also reported disruption. Google has promised to publish an analysis.

It is not the first time this region has gone down. A cooling failure during the 2022 heatwave and a loss of power in August 2024 both hit Google's London cloud.

The outlets were Reuters, AP, the Guardian, the Wall Street Journal, Business Insider, ITV News, Sky News, The Register, TechRadar, Computer Weekly, Tom's Guide, Central Banking and Anadolu Agency. That is three wire services, British and American newspapers and broadcasters, and technology and finance specialists. Copies of wire stories run by other outlets were left out, so no report is counted twice. The Wall Street Journal and Central Banking sit behind paywalls, so only the opening of each was scanned. Sky News, TechRadar and Tom's Guide ran live blogs.

The result

All thirteen read as human. None drew a single writing-habit flag.

ITV News and Business Insider were the least decisive, and both still read clearly as human.

This is the second story scanned with Brath, Fíor's new model. In the first, Meta's Muse Wonder, twelve of thirteen outlets read as human.

Why every one read as human

Outage coverage is built from particular things. A power failure in one named zone. A list of banks. A driver who could not take card payments. A passenger who could not buy food on a train. Several outlets updated live through the day as Google posted fixes.

Specific detail is the obvious explanation. It does not settle it. The Register and Computer Weekly wrote technical pieces about cloud zones and backup systems, with few people quoted in them, and they read just as clearly as human.

There is another possible explanation, and it belongs in the open. Brath was built to be less harsh than the model before it. Two stories in, nearly everything it has scanned has read as human. That may say something about how this news was written. It may also say something about the model. Two stories cannot separate those, and Fíor cannot tell them apart on its own. The next few stories will be read with that in mind.

When one zone of one London region lost power, people across Britain could not pay for their lunch. Every account of that, from the wire reports to the live blogs, read as if a person wrote it.

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.