7AI - The Agentic Security Platform - Blog

AI played a role in the recent outages, just not the one most people fear

Written by Yonatan Striem Amit | Nov 20, 2026, 1:17:32 PM

Over the past week, an air traffic control failure grounded flights nationwide, Microsoft 365 went down, and customers at major banks couldn't reach their accounts. Unsurprisingly, a lot of people started asking whether AI was behind it, either directed by attackers or acting on its own.

I got that question directly from Boston Globe technology columnist Hiawatha Bray, whose column concludes the outages were most likely a coincidence. He's right that there's no evidence of a coordinated campaign or a rogue AI behind these failures, and right that our digital infrastructure is far more fragile than most people realize.

But I think AI is part of the story, just not the way most people worry. The connection runs through the defenders: AI is finding weaknesses faster than organizations can safely fix them. It's also worth being clear about what documented AI-driven attacks actually look like, because that's how you tell whether you're looking at one.

How AI connects to these outages

AI is exposing vulnerabilities at a rate we've never seen before, creating the biggest attack surface in the history of cyber. The time it takes to develop an exploit is now negative: Mandiant's M-Trends 2026 puts the mean time to exploit at negative seven days, meaning attackers routinely exploit before a patch exists. The old mechanisms are being stretched beyond the point of breaking.

So defenders are scrambling to patch as quickly as possible, and that regrettably means skipping best practices like thorough testing and gradual rollout. That causes breakage. The Washington Sun reported that bank executives and the NSA have told members of Congress this is what's behind the recent outages. The sourcing is anonymous, and publicly the companies have pointed to bugs and maintenance, so I'd treat it as early. But it fits what we'd expect.

These defenders are right to worry, and right to move fast, even at the risk of disruptions like these. We need to start having a more mature conversation that weighs the impact of a disruption against the risk of a cyberattack that does dramatically more damage. A grounded flight is expensive and miserable. A successful attack on the same systems could be far worse.

A week like this is a gift to scammers

Whatever caused these outages, attackers will take advantage of the moment. Millions of people just had flights canceled, couldn't get into their email, or couldn't reach their bank accounts. Every one of them is primed to believe a message about it.

Voice attacks are especially effective right now, because a phone call about an outage everyone has heard of sounds plausible. And expect groups to claim credit for outages they had nothing to do with. A false claim costs nothing and gets attention. Be skeptical of any message or call this week asking you to log in, verify an account, or confirm payment details.

What documented AI-driven attacks look like

AI attackers are here, and the evidence gives us a clear picture of how they behave. The patterns aren't that different from regular attacks: reconnaissance, exploitation, stealing credentials, moving toward data and access. Palo Alto Networks' Unit 42 reached the same conclusion: these tools target long-standing weak points like exposed APIs and stored credentials. What changes is that they're accelerated and exploratory. Imagine a large, gifted team of hackers going after every route, all the time, in parallel.

ARTEX, the autonomous attack framework the FBI warned about in a FLASH advisory last week, works this way. Once someone sets a goal, AI agents find and exploit weaknesses with little human direction, and when one path fails they try the next. The FBI says some intrusions went from initial targeting to privileged access in under three hours, and BleepingComputer has tracked the campaigns tied to it. That’s intentional impact, and it leaves a trail investigators recognize. Our threat research team ran a simulated ARTEX intrusion to show exactly what that trail looks like: What ARTEX teaches defenders about autonomous AI attacks.

The accidental cases have a pattern too. When OpenAI's models broke into Hugging Face during internal testing, they were pursuing an ordinary goal by an unexpected route, and it stayed quiet for weeks. As I told Hiawatha, if AI misbehaves at scale, I'd expect it to look like normal activity that's slightly wrong.

In both cases there's a goal you can trace. So when investigators look at an outage, randomness by itself doesn't point to AI. What they look for is agent activity and the objective it was working toward. Nothing reported about these outages fits either pattern.

Getting ready for next time

Hiawatha's column ends with a warning: next time we might not be so lucky, so we'd better get ready. I agree. Getting ready means a few concrete things:

  1. Shrink what attackers can find. The FBI's top recommendation on ARTEX is to inventory and reduce your external attack surface, because autonomous tools can't attack what they can't see.
  2. Make remediation as fast and as safe as discovery. Prioritize what's actually exploitable, and invest in the testing and rollout capacity to patch quickly without breaking production.
  3. Watch for opportunists during every disruption, whatever its cause.
  4. Fight AI with AI, with humans on the loop making the decisions that matter.

We put together a full field guide on getting ready for autonomous attacks, using ARTEX as the case study: Autonomous AI attacks are here.

As an industry, we're still learning how to fight AI with AI. Weeks like this one are a reminder that we need to learn fast.

Frequently asked questions

Did AI cause the November 2026 outages?

There is no evidence that malicious or rogue AI caused the recent outages. Microsoft, the FAA, and the affected banks have pointed to technical causes such as bugs and maintenance. The Washington Sun reported that officials told Congress the outages stem from rushed patches for vulnerabilities that AI models found, though that reporting relies on anonymous sources.

How can AI contribute to outages without an attack?

AI is finding vulnerabilities faster than organizations can safely fix them. Mandiant's M-Trends 2026 puts the mean time to exploit at negative seven days, so defenders are under pressure to patch immediately. Rushing patches means skipping thorough testing and gradual rollout, and that can break systems.

What scams should people watch for after a major outage?

Expect airline refund phishing, fake "please sign back in" emails, calls and texts claiming to be your bank's fraud team, and callers posing as IT support who are "fixing the outage." Groups may also falsely claim credit for outages they didn't cause. Don't use the link or phone number in the message. Go to the company directly.

What does an AI-driven cyberattack look like?

AI-driven attacks follow the same patterns as conventional ones, including reconnaissance, exploitation, credential theft, and moving toward data and access. The difference is speed and breadth: AI agents try many routes in parallel and switch paths when one fails. The FBI says some intrusions using the ARTEX framework reached privileged access in under three hours.

How can organizations prepare for AI-driven attacks?

Start by shrinking the external attack surface, which is the FBI's top recommendation on ARTEX. Make remediation as fast and as safe as discovery, and watch for opportunistic attacks during every disruption. 7AI's view is that defenders need to fight AI with AI, with humans on the loop making the decisions that matter.