Last week was operational hell for the nation’s key systems…

Within days, Microsoft Outlook and Teams went down, a nationwide FAA ground stop held flights across the country, United had a systemwide outage, banking apps failed, and the NSA's internal classified network had problems.

The obvious question was whether this was a coordinated cyberattack. It wasn’t!

I have spent my whole career in security and I want to push back on the conclusion many people will draw which is: not an attack, so not a security problem!

1. Resiliency is a key part of security

Security has always rested on three pillars: confidentiality, integrity and availability. Too often we treat the third as someone else's job, the infrastructure team or the cloud provider. To a traveler stuck at the gate or a customer who can't reach their bank account, a nation-state attack and a botched router upgrade look exactly the same. The service is down either way.

The FAA ground stop demonstrates this. It appears to trace back to an upgrade in Arkansas that went wrong, and a single local mistake took down the entire national system. A single point of failure is a vulnerability, whether or not an attacker finds it. Redundancy, failover, rollback and tested recovery aren't just operational hygiene but also security controls. We should design, measure and audit them that way.

2. AI has changed the speed of patching

As Dmitri Alperovitch put it starkly:

“More vulnerabilities have been discovered this year than the last 30 years combined.”

AI models are finding flaws in software faster than ever before. Enterprises and government agencies are flooded with critical vulnerability reports. He called it a “patch apocalypse,” and he expects the next 12 to 18 months to be very rough.

This is where the two points intersect. Every patch is a change, and change is the most common cause of outages. When organizations speed up their patching without strengthening their resilience, they get what we saw last week: systems taken down by their own updates rather than by an attacker.

Fast patching and resilience are therefore one problem, not two. You can only patch at the speed AI now requires if you're confident you can survive a bad update.

The one takeaway for companies

Test your resiliency not just have a resiliency plan.

Most organizations have a disaster recovery document somewhere. Far fewer have actually pulled the plug and watched what happens.

  • Roll back a production update on purpose and time how long it takes.
  • Fail over your critical systems and verify they actually recover.
  • Run a day without your primary collaboration tool or identity provider.
  • Map your real dependencies, including your vendors' dependencies, and find the hidden single points of failure.

If you haven't tested it, you can't rely on it.

The one takeaway for end users

Don't do anything in haste.

Things can be built for resilience and still fall to a single click. When services are down people get frustrated and eager which scammers take advantage with messages like:

  • Recent Microsoft outage, please re-verify your password here
  • Your flight was canceled, click to claim your refund
  • Confirm your bank account details to restore your access

So my advice to end users is to not act in haste, take a deep breath.

Last week wasn’t an attack but a warning. AI is forcing everyone to patch faster than ever and only organizations who have tested their ability to absorb failure can do it safely.

If you are an organization, treat resiliency as a key security control and if you are an end user, don’t do anything in haste!