The backup is a phone

At 2:42 a.m. on February 22, 2024, AT&T made a change to its wireless network that contained a configuration error. Three minutes later every device on the network was disconnected, nationwide. It took close to two hours to roll the change back, and at least twelve to restore full service, because the systems that register phones onto the network were overwhelmed by everything trying to reconnect at once. By the Federal Communications Commission's count the outage affected more than 125 million devices and blocked more than 92 million calls, more than 25,000 of them to 911.¹

One error, one company, every phone on its network at the same moment. That is the shape of the scenario behind this piece, whose title is four words long: "Apocalypse: Apple Phones." What if every iPhone stopped working tomorrow?

The rules of the game

It is a thought experiment, so it needs rules. Say every iPhone stops working at once, and nothing else does. Macs, iPads, Android phones, websites and networks carry on. The question is not how likely this is. It is what the exercise reveals about how things are connected.

The scale is not small. By one measure, share of mobile web traffic in the United States, Apple devices accounted for about 54 percent in September 2026.² That counts page views, not phones, so it is a proxy, and the share of devices could differ. It is enough to say that the scenario takes out a very large part of the country's smartphones.

Many things that used to fail one at a time

A wallet, house keys, a map, a camera, a boarding pass, a bank's security token, a driver's license, an alarm clock and a telephone used to be separate objects. Losing your keys did not cost you your map. Failures were independent, and independence was a kind of safety nobody had to pay for.

The phone merged them. Then a large share of those phones came from one company, running one operating system, updated through one channel. Failures that used to be independent became correlated.

That much is a familiar observation. The real outages show it is not hypothetical. In July 2024 a single faulty update from the security company CrowdStrike crashed about 8.5 million Windows machines, by Microsoft's estimate less than one percent of them. Microsoft noted that the "broad economic and societal impacts" came from who was using them: enterprises "that run many critical services."³ Share is not the point. What matters is which machines are in the failing set.

What fails first

In the scenario, the first things to go would probably not be conversations. People would find a laptop, a landline, a neighbor. What would fail fast is proof: the ability to show who you are, pay, log in, or get through a door, because those increasingly run through a code or a key that lives on the phone.

This is where the scenario stops being about a device.

The backups are made of phones

Apple has thought hard about people losing their phones. Its support documentation says passkeys, the password replacements stored in its keychain, sync across a person's devices to provide "redundancy in case of loss of a single device," and that they can be recovered "even in the event that all associated devices are lost," from an encrypted copy Apple holds but cannot read.⁴

Look at how that recovery works. To get the keychain back, a user must sign in with their account password, "respond to an SMS sent to their registered phone number," and enter their device passcode.⁴ Apple also lets people name a recovery contact, a friend or relative who can generate a code from their own device to help them back into their account.⁵

All of that is sensible for the case it was built for: you lose your phone, and everything else is still standing. Now run it under the scenario. The text message goes to a phone number, and for iPhone 14 models sold in the United States, the number lives in an eSIM inside the phone, with no physical SIM card to move to another handset.⁶ The recovery contact is a person who may quite possibly be holding a dead iPhone too.

The pattern is the interesting part. Personal redundancy, the backups for one person's phone, is built largely out of other phones and other accounts on the same platform. That protects against losing a phone. It does not protect against all of them failing together, because the backup and the thing being backed up go down at the same moment.

Nothing here says Apple designed badly. A recovery system for correlated failure would be a different and stranger thing, and nobody asked for one. The point is that the scenario shows something the design document is not meant to.

Coming back is correlated too

The AT&T outage carries a second lesson. The error took about two hours to undo. The rest of the twelve hours went to the stampede: every disconnected device trying to rejoin the network at once.¹

A system that fails all at once also tries to recover all at once. Every recovery path sized for the ordinary trickle of lost phones, support lines, account recovery queues, replacement stock in stores, would meet the whole population on the same morning. The time to recover from a correlated failure is set less by the fix than by the queue.

The strongest objection

The obvious objection is that this will not happen. Every iPhone failing at once is extraordinarily unlikely, and a thought experiment about an event with no history can make any risk look large.

Fair, and the article does not claim otherwise. But the exercise does not depend on the odds. Partial versions already happen: one network, one update.¹³ Each one runs a small piece of the experiment for real. The scenario just asks what the full map looks like.

A second objection is that people would cope: borrow phones, visit branches, wait it out. Many would. Coping depends on fallbacks, and the fallbacks are what has been thinning, because a path almost nobody uses is a path that costs money to keep.

An open edge

Independence used to come for free. Because tools were separate objects, they failed separately, and nobody had to plan for that. Consolidation spent that independence without anyone deciding to spend it, and there is no one whose job is to count what is left.

Each bank, airline and agency sees only its own reliance on phones. The shared dependency sits between all of them, on nobody's map. The author's four-word title is a way of drawing that map in one stroke. A smaller version is available to anyone: list what you could not do tomorrow without your phone, and then ask, for each one, what the second way used to be.