Skip to main content
← Back to Source Journal

The Instrument

How a Wave of Outages Could Make the Case for Blockchain Rails

20 November 2026 · Ben Visser · 4 min read

At a glance - TL;DR

In five days in November, Microsoft 365, the FAA and Chase all went down, and each time the official word was not cyber and not AI. Ben reads the three explanations side by side and finds the same thing underneath: change landing on old systems that cannot take it, driven by AI finding software flaws faster than anyone can patch them. He follows that chain forward, through one large event or an accumulation of small ones, to a trough and a narrative that makes shared ledgers run by agents feel necessary, while keeping the risks of the new rails in view.

Key takeaways

  • Three major outages in five days were each officially not cyber and not AI, yet all three describe the same thing: something changed, and something old could not take it.
  • AI is finding software flaws faster than they can be fixed: Oracle patched 1,449 vulnerabilities in July against 309 a year earlier.
  • The pressure on legacy systems is less a wave of attacks than a wave of change.
  • Nobody has to invent the story that we need safer rails; the institutions with the next system already built will be ready to tell it, and it will be largely true.
  • A ledger can record who gave an agent its permission, and when. It cannot record why.

Hey there,

Last week I wrote about the half of the AI story the big podcasts leave out: the blockchain rails being built underneath, and what happens when agents start to run on them. Since then I have caught myself looking further ahead than I usually do here, and this week gave me a reason to.

Five Days in November

On Thursday the twelfth, Microsoft 365 went down across much of the world for about a day. On Friday the thirteenth, the FAA stopped every departure in the United States for four and a half hours, after communications failed at 17 of its 22 en route centres. On Monday the sixteenth, Chase customers could not see their balances, and because the big American banks share payment rails, delays spread to Wells Fargo, Bank of America and Zelle.

Each time the first official word was the same: not cyber, and not AI. Microsoft traced its outage to a hidden fault from an earlier network upgrade, exposed by a later change. The FAA pointed to a router replacement that overloaded antiquated equipment, though its own administrator said that did not fully explain the root cause, and the contractor disputes the blame. Chase called it "a software bug in an on-premise hardware system".

I believe them. Nothing I have read suggests an attack. But reading the three explanations side by side, I kept noticing that they describe the same thing. Something changed, and something old could not take it.

Why the Old Systems Are Shaking

There is a reason the old systems are being asked to change so often right now, and it is AI.

This year AI started finding software flaws faster than people can fix them. Oracle's July update patched 1,449 vulnerabilities, against 309 in the same month a year earlier, and Microsoft set its own record in the same week. Both credited AI-assisted discovery. The American vulnerability database logged more than 45,000 flaws in seven months, and in April NIST stopped fully processing new ones and shelved about 29,000. Meanwhile the average time to fix a critical flaw in complex enterprise software is about five months.

Most of these flaws are never exploited. The pressure is not a wave of attacks. It is a wave of change, landing on banks, airlines and agencies that run on software older than some of the people patching it. Every patch, upgrade and router swap is a small risk, and there are more of them each month.

What Is on the Shelf

At the same time, the replacement is being built, mostly in public. In July the Bank for International Settlements had twenty-eight institutions settle real money on Project Agorá, one shared ledger for tokenised bank deposits and central bank reserves. DTCC and The Clearing House are aiming at the first half of 2027 for tokenised securities and deposits. Europe wants the digital euro law agreed by the end of this year.

It is easy to see why this appeals. Today a payment crosses many old systems, each patched separately, each able to fail on its own. A shared ledger replaces that with one record that every party can check. Agents can run the work around it, reconciling, monitoring and rolling out fixes in batches, at a pace no team of people could match. When old systems keep falling over, that starts to sound less like an experiment and more like a relief.

If This, Then This

So here is the chain I keep running.

If AI keeps finding flaws faster than old systems can absorb the fixes, and the outages keep coming, then the question is how they arrive.

One way is a single large event. A failure that does not clear in four hours, or one that lands on the payment rails during a moment of financial stress. I think of the Sunday evening in August 1971, when Richard Nixon ended the dollar's link to gold on television in about twenty minutes. Big switches tend to be prepared slowly and made fast, announced as a rescue, with the new system already on the shelf.

The other way is quieter. No single event, just an accumulation: a bank on Monday, an airline on Friday, a hospital the week after. Each one is explained and fixed, and each one wears away a little more patience with the way things work now.

Either way, I expect a narrative to form: we need safer, authentic infrastructure, built on shared ledgers and run by agents, to weather this storm. Nobody has to invent that story for it to appear. The institutions that already have the next system built will simply be ready to tell it when people want to hear it. And the story will be largely true, which is what will make it so persuasive.

There is a third path I should leave standing. The old systems hold, the patching catches up, and the switch happens slowly over a decade without any storm at all. I do not think that is the most likely path. But I could be wrong, and I want this entry to leave room for it.

The Trough

If the storm comes, I suspect we go through a deep trough before we see the better side of it. Outages cost people their flights, their wages on the day they are due and their trust in systems they never had to think about. A society that keeps being let down that way gets tense, and tense societies accept changes quickly, along with conditions they might have argued about in calmer times.

This is where I try to keep both hands of the instrument in view. The new rails are also software. Ledgers can be sound while the bridges between them are hacked, and agents that patch in batches can push a bad fix in batches too. Programmable money brought in during a crisis tends to arrive with its strictest settings switched on, and whoever writes those settings will shape how agents act for years.

What the Storm Cannot Decide

Whichever way it goes, one thing stays the same. Every agent on these rails will act on a permission someone wrote down: how much, for what, until when. The ledger will record who granted it and when, and nobody will be able to quietly change it afterwards. What it cannot record is why. That stays with a person, in calm weather and in a storm.

It is late as I finish this, and the house is quiet. Somewhere a router is being swapped and a patch is being tested, and most nights nothing happens. I do not know which way the weather turns. I would like us to know our reasons before it does.

With care, Ben