On Monday 16 November 2026, a software release that JPMorgan Chase pushed to its own on-premise systems that morning slowed transactions for most of the business day. Customers at other banks felt it too, because so much money moves through Chase. It was not an attack. It was a change, and that makes it far more useful to learn from.

Read this first. Facts as of 17 November 2026. Chase has not published a technical post-mortem. I tag claims as [Official] (on record), [Reported] (press, including anonymous sources) and [Inferred] (my own reasoning).

What happened

  • The trigger. An FBI official said the disruption was traced to a software release the bank rolled out earlier Monday that affected a subset of its systems, with no sign of malicious or external activity. [Reported]
  • Where it ran. Chase called it “a software bug in on-premise hardware systems.” A source told CNN it was not cloud-related: an update on the bank’s own physical systems caused it. Another source told NBC it was confined to retail banking. [Official + Reported]
  • When. Downdetector reports began around 9:30 AM Eastern and spiked about two and a half hours later. A teller at a Washington, D.C. branch said problems started around 10 AM and were resolved just after 3 PM. Chase said transactions were posting normally by evening. [Reported + Official]
  • Who did it. Chase’s own change, on Chase’s own hardware. Nobody has said who approved it, what component it touched, or why it went out on a Monday morning. [Inferred]

What broke, and what didn’t

This was a degradation, not a blackout. Chase said “some transactions are taking longer than usual to complete but customers still have access to their accounts and funds.” [Official] ATMs largely kept working, while branch systems were “up and down”, according to staff. [Reported] JPMorgan shares still fell more than 2%. [Reported]

The Downdetector numbers

Service Reports (peak, approx.) What users reported
Chase ~4,700 by 2 PM ET; nearly 5,000 at peak About half about balances and transactions
Bank of America Just over 1,000 Access and transfers
Wells Fargo 800+ Logins, balances, fund transfers
U.S. Bank Hundreds Zelle delays
Zelle Hundreds Payments not arriving
Capital One, Truist Smaller spikes Access

Sources: CNN, Al Jazeera, Gizmodo, Tom’s Guide, CBS News. Downdetector counts user reports, not confirmed failures. [Reported]

Why did other banks’ customers complain?

This is the puzzling part. Bank of America said “our systems have been operating normally today”; Wells Fargo and PNC said they were not impacted; U.S. Bank said its systems were “not directly impacted” but Zelle payments with other banks could be delayed. Zelle said it was “not down” and that the issue was at a financial institution. [Official] Yet over a thousand Bank of America customers and 800 Wells Fargo customers complained, some about logins and balances, not just transfers. Two explanations fit:

1. Perception, not failure (most likely)

Downdetector is crowd-reported. A Bank of America customer whose Zelle payment to a Chase account hangs reports “Bank of America is down”, because that’s the app in their hand. CNN noted the other banks’ reports appeared related to sending money through Zelle to Chase accounts. [Reported] Then the news amplifies it: once headlines say “banks are down”, people open their apps to check, and any ordinary delay becomes a report. The order of magnitude supports this: about 5,000 for Chase against hundreds to a thousand for everyone else. [Inferred]

2. Real technical coupling (hypothesis, not confirmed)

Modern banking apps build their home screen from several back-end calls: balances, recent activity, pending payments, Zelle status. If one of those calls waits synchronously on a slow counterparty, without a short timeout, the whole screen can feel slow or fail to load, even though the bank’s core is healthy. Pending transfers to Chase that never confirm could also show odd balances. No bank has confirmed this, but it is a well-known failure pattern, and it is exactly what timeouts and circuit breakers are for. [Inferred]

We have seen this before

  • Barclays, January 2025. A software problem in a critical module of its UK mainframe operating system “caused progressively severe degradation of mainframe processing performance”. Over three days, 56% of online payments failed. The fix came later from the mainframe vendor. [Official]
  • UK banks, 2023–2025. The UK Treasury Committee counted more than 800 hours of IT outages at major banks in two years. Common causes: third-party suppliers, disruption from a change in systems, and internal software malfunctions. [Official]
  • TSB, 2018. A platform migration went live without enough testing evidence and locked customers out for weeks; the board had approved go-live. [Reported]
  • CrowdStrike, 2024. One faulty update crashed Windows machines worldwide, banks included, because it reached everything at once. [Reported]

The pattern: the dangerous moment in banking IT is rarely the attacker. It is the change. [Inferred]

Technical lessons learned

1. Watch for slowness, not just errors

Nothing was “down”, so health checks probably stayed green while transactions queued. Barclays was also a progressive slowdown. Alert on latency percentiles, queue depth, pending transactions and settlement lag, and compare them with the same hour last week.

2. Canary with real traffic, and roll back automatically

A release that slows a subset of systems should be caught on a small slice before it reaches everyone. Gate each rollout stage on the business metrics above, and let the pipeline roll back on its own when a threshold is crossed, without waiting for a human to read Downdetector.

3. On-premise needs a real way back

In the cloud you swap instances. On your own hardware, keep the previous version installed and ready (blue/green, A/B boot partitions, or a parked old binary) so rollback is minutes, not a rebuild. Test that path before every release, not during the incident.

4. Contain the blast radius for others

Chase is a hub. Its slowness became other banks’ problem through Zelle. Downstream, use short timeouts, circuit breakers and bulkheads, and build screens so one slow dependency degrades one widget, not the whole app. Upstream, if you are the hub, your change risk includes your partners’ customers.

5. Don’t ship into peak hours

A Monday morning release on a payments path is a choice. Banking IT people say banks avoid changes at month end; the same logic applies to any peak. Use low-traffic windows, freezes around paydays and holidays, and treat exceptions as high risk.

6. Shared code is a shared failure

A release that hits many systems at once is the CrowdStrike lesson again. Separate failure domains (by region, by product line, by host group) and never update all of them in the same wave.

7. Communicate early and precisely

Chase did one thing well: “technical issue, not cyber, funds safe” came out the same day and calmed a week of attack rumors. The other banks helped too by saying clearly that their systems were fine. Pair that with a public status page and a post-mortem later.

A quick checklist

  • ☐ Release tested on production-like hardware, not just in containers.
  • ☐ Canary slice with business-metric gates: latency, queue depth, pending transactions.
  • ☐ Automatic rollback wired to those gates; previous version staged and tested.
  • ☐ Window outside peak hours, paydays and month end.
  • ☐ Failure domains updated in separate waves.
  • ☐ Short timeouts and circuit breakers toward every counterparty; UI degrades per widget.
  • ☐ Status page and customer message templates ready before the change.

Open questions

  • What component did the release change, and why did on-premise hardware misbehave?
  • Was there a canary, and did it miss the problem?
  • Why did recovery take about five hours, and was it a rollback or a forward fix?
  • Did any other bank see real slowness in its own systems from waiting on Chase?

Sources

Facts as of 17 November 2026. Chase has not published a post-mortem, and some details here may change.

Tags: , , , , , ,

Leave a Reply

Your email address will not be published. Required fields are marked *