When routine changes bring down a nation

Egil Østhus

Egil Østhus

CEO & co-founder

November 20, 2026

On Monday, Chase customers across the US watched their transactions stall. The FBI looked into it and found no sign of an attack. The cause simple was a software release Chase had shipped that same morning.

It was that kind of week. Microsoft took down Outlook and Teams for more than a day with a network capacity upgrade. The FAA grounded departures nationwide after a contractor installed a new router. Apple blamed an internal update for an outage that hit iCloud, the App Store and Apple Music. By Wednesday the Vice President of the United States had canceled a speech to sit in meetings about it. The week before, the CEO of Westpac, an Australian bank with branches in New York, London and Singapore, was taking questions in parliament two days after his customers lost their banking app for 12 hours.

These are not careless organizations. They run change boards, release windows and rollback plans. Those controls were built for a time when a bank shipped a few big releases a year and a regulator mostly asked whether the system was up. That time is over.

Your teams ship more code with AI. Decide what reaches users.
DORA research links AI coding agents to lower delivery stability. Unleash lets you release gradually and roll back any feature instantly.Book a demo

Change got fast, and the controls stayed slow

More than 90% of developers now use AI every day, and code reaches production faster than any review process was designed for. BCG found that 73% of major banks already let AI agents run in production. While only 8% have a comprehensive way to control them. About a third of financial services outages already trace back to a software deployment. And each hour costs around $1.7M, well past the $400,000 often attributed to IBM.

That meter runs by the hour, and the hours pile up because finding the fault is the quick part. The fix still has to be rebuilt, approved and pushed out to every system before a single customer sees it. Until then, payments keep failing and call centers keep backing up. That’s how a bad morning becomes a lost day.

That is before regulatory fines. For example, UK regulators fined TSB £48.65M for a failed IT migration, plus £32.7M in customer redress.

Regulators have noticed

In the US, the OCC’s new examiner framework for AI in banks expects a bank to prove it can stop a system that is running but doing the wrong thing, switch to a backup and get back to normal, and that it has rehearsed all of it. In Europe, DORA requires financial firms to manage ICT changes in a controlled way, with fallback procedures. UK rules hold banks and insurers to impact tolerances for their important services. They all want the same evidence, which is who approved a change and how fast it was reversed. A change board can produce that, but only by slowing everyone down.

You break it, you bought it

This week showed that a routine change can take down an entire system, and change is now shipping faster than ever. The controls most companies rely on to manage change were designed for a slower pace and for older rules. They fit neither today’s technology nor today’s regulation.

That makes this a C-level problem, not an IT problem.

The CEOs, CIOs and CTOs of major financial institutions now clearly own the risk of change. When a release goes wrong, they are the ones who end up in front of a regulator or a parliament, so they are the ones who have to enforce new controls. But they need to do it without slowing their engineering teams down, because a bank that ships slower loses to one that doesn’t.

In an AI native world, that means the C-Suite needs to enable IT with people, process and tooling to adopt AI at full speed without losing control while driving the risk of any single change toward zero.

Three new controls for engineering teams

For engineering, that comes down to three new controls.

  • Gradual rollout, a kill switch and instant rollback for application behavior, so a bad change is switched off in seconds instead of repaired over hours.
  • Change management that is audit-ready by default, so every change is approved and recorded without a change board standing in the way.
  • Control of AI-written code, so when one change misbehaves, you know which one it was and can turn off just that one.

That’s what we call FeatureOps: runtime control for applications that lets teams ship faster while governance gets stronger. It’s what we build at Unleash, and Lloyds Banking Group, Prudential and Visa already use it.

If you own the risk of change at your bank, give us 30 days and one revenue-critical service. We’ll put its releases behind a kill switch, run a game day with an injected fault, and show you a failure contained in under a minute. You can’t afford to wait any longer. Not when a routine change can bring down a nation.

Explore further