Home / News / Article
SecurityWeek · 14 min read

Industry Reactions to Major Outages and AI Patching Questions: Feedback Friday

A series of major technology outages have hit the United States, disrupting email, air travel, banking and government operations.

The incidents included a nearly day-long Microsoft 365 outage, a nationwide FAA ground stop, a technical issue at JPMorgan Chase that slowed transactions at other banks, and an outage affecting the NSA’s classified internal network. United Airlines and Apple also experienced disruptions.

Officials said there was no evidence of a cyberattack or a link between the incidents. Microsoft traced its outage to a months-old software defect, the FAA blamed a newly installed router, and the NSA cited routine maintenance.

Director of National Intelligence Jay Clayton, however, linked the NSA disruption to patching for a PHP vulnerability found with the help of AI. Sen. Mark Warner said AI-driven vulnerability discovery is pushing organizations to patch many systems at once, putting strain on infrastructure that wasn’t built for that pace of change.

Industry professionals have commented on whether AI-accelerated patching played a role, the need for tested rollback plans, and the fragility of interconnected systems.

And the feedback begins…

Gil Tene, Chief Technology Officer & Co-founder, Azul:

“We have strong reason to believe that AI-accelerated patch velocity is already causing major operational incidents. Based on direct experience with customers, we know that urgently deploying large, non-security-only OpenJDK updates does cause stability incidents. These rollouts sometimes fail and have to be reverted, leaving organizations exposed and delaying the remediation of the very vulnerabilities they are meant to address until teams find a workaround they can roll out without compromising stability.

As AI accelerates the discovery of critical flaws, patch cycles are compressing rapidly. In the past four weeks alone, the Java community has had to absorb four update events, including three separate critical vulnerability events and a multitude of high-severity CVEs that couldn’t wait. When a critical Java vulnerability lands, teams must push OpenJDK updates into production within hours, often with less testing than they would normally do and with far more urgency to get the change rolled out before attackers can exploit it. When the fix for a vulnerability is available only as part of a much wider full update that brings along hundreds of unrelated changes, the rollout carries much more risk, because change is risk.

The outage risk doesn’t come from the vulnerabilities themselves. It comes from the volume of change forced by compressed timelines. Hurried, urgent change is risky to begin with. The more change a hurried rollout carries, the greater the risk of instability and widespread incidents. Security-only updates minimize that risk by limiting the change to what is necessary to resolve the vulnerabilities, and nothing more.”

Samuel Hassine, CEO and Co-founder, Filigran:

“I think [faster patching pace driven by AI] plays a part, but I would be careful not to make it the whole story. From what has been made public so far, these outages look less like an AI problem than a change problem: updates and upgrades pushed onto systems that can’t take much change. AI is making that visible, because the list of flaws to fix now grows faster than most companies can safely roll out the fixes.

For years our industry has told companies to patch faster and has measured them on that alone. But a fix that takes the service down is a security failure too, because being available is the first thing a customer expects from you. So the question I would put to every CEO is who owns the choice between leaving a flaw open a few more days and risking the service by closing it tonight. In most companies those are two teams with two different targets, and at this pace it has to be one person who answers for both.”

Ken Barth, CEO, Catalogic Software:

“AI is finding vulnerabilities faster than many remediation processes were built to handle. Microsoft fixed more than 1,400 vulnerabilities in its October security release, a monthly record, and 1,094 in November. The company has said AI-assisted discovery will put more fixes into each release. Each fix is a change to a live system. With the same staff and maintenance windows, a higher volume leaves less time to test each update, and less testing leaves more room for something to break. Nothing published so far links this pressure to the recent Microsoft 365 outage, but Microsoft’s own explanation shows how easily planned work can take a service down.

Waiting longer is not a safe answer either. Google has reported attackers using AI to reverse-engineer patches and quickly build exploits for the flaws the patches address, so every published fix also tells attackers where to look. That makes being able to undo a change a security control in its own right. Before it reaches production, each update needs a verified restore point for the systems it touches, a tested rollback plan, and a clear owner for the decision to use it.

The same discipline applies after something breaks. Microsoft restored services in stages to keep them stable, which is what careful recovery looks like. Organizations should know before the next incident which services must come back first, what each one depends on, and how long the full sequence takes. A recovery time that has never been tested is only an estimate, and boards should treat it as one.”

Brian Contos, Field CISO, Mitiga:

“Faster patching isn’t the problem by itself. AI is shrinking the time between a vulnerability being disclosed and being exploited, so defenders are under real pressure to ship fixes faster, and every change is a chance to break something. Look at the past week. Officials tied the FAA disruption to a network equipment upgrade, sources told CBS News the NSA outage followed routine patching, and JPMorgan Chase attributed its problems to a software issue. None of those appear to be attacks. What turns a routine change into a national outage is pushing it without being able to see its effects or quickly undo it. Speed is fine if every change is staged, watched for how the system behaves afterward and backed by a rollback someone has tested.

The other lesson is how far a single failure travels. One router change at the FAA destabilized 17 of its 22 en route centers, and Chase said its issue may have slowed transactions at other banks. Redundancy alone doesn’t protect against a healthy component that’s misbehaving, because everything around it trusts it. Segmentation, independent failure domains and a rehearsed process for isolating a bad component quickly are what keep one problem from becoming everyone’s problem.

The preparation for failures and cyberattacks is largely the same. An adversary can deliberately create these same disruptions, coordinate them across multiple locations and interfere with recovery. If a system can’t survive an accident, it won’t survive an adversary.”

Benny Czarny, CEO and Founder, OPSWAT:

“Faster patching can actually contribute to outages when testing and recovery cannot keep up. AI is increasing the pressure to fix vulnerabilities quickly, but you can’t attribute anything to AI without evidence from the investigation.

Failures this week have resulted from cloning systems and their weaknesses along with them. When a primary system and its backup run the same software, receive the same update, and depend on the same network, then they really aren’t that independent.

From a cybersecurity standpoint, critical infrastructure needs the same design mindset as an airplane cockpit, where primary and secondary systems use different hardware and software to reduce the chance that one defect disables every way of performing a critical function.”

Bryson Byrd, Cybersecurity Advisor, Huntress:

“Right now, there is no definitive link between these outages, and we should be careful about rushing to conclusions that AI or cyberattacks are responsible without having all the facts. What we do know is that there is a significant AI race going on between defenders and threat actors. As AI accelerates vulnerability discovery, defenders are under increasing pressure to engineer, test and push patches quickly to stay one step ahead. When companies are focused on speed of development, mistakes can happen. In the age of rapid patching, testing and verifying updates is arguably more important than ever.

My concern is that these outages will cause organizations to start dialing back patches out of fear of being the next Microsoft or FAA. We cannot afford to go back to a time when system administrators could wait several days before deploying critical security updates. The window between vulnerability discovery and potential exploitation is shrinking, and slowing down patches could leave organizations exposed to threats they already know about. So we cannot stop updates or sacrifice testing for speed. We need to double down on resilience and make sure organizations are prepared when an update doesn’t go as planned.

When it comes to critical patches and updates, there should always be a fallback plan in place, whether that’s a known, good backup or the last working code push that teams can quickly return to. We often see security teams with these plans documented, but the mechanisms haven’t actually been tested and verified to work. As infrastructure continues to evolve, these plans need to be tested regularly. Otherwise, when a bad update happens, organizations may not be prepared to respond, and a relatively contained issue can quickly grow into a much larger disruption.”

Grayson North, Principal Threat Intelligence Consultant, GuidePoint Security:

“This situation is a combination of increased media attention and a flurry of technical changes caused by AI’s rapid acceleration of vulnerability discovery. Outages are fairly common, but a few hours of downtime on a highly-visible system draws heightened attention. It can be compared to the DCA airline crash last year in a sense. After that incident, every near miss got national attention rather than being the usual footnote.

We are also clearly in a period of flux in regards to vulnerability discovery/patch management. AI has made discovery of vulnerabilities trivial for those willing to spend tokens. As such the pure number of vulnerabilities being reported is through the roof (thus the “Vulnpocalypse” that some have dubbed the moment). Some of these discoveries lead to big, scary cyberattacks, yes, but all of them lead to patches to fix the problem. Every time you make a change to a production system, you risk causing an unforeseen issue. This risk scales exponentially with the number of patches and number of underlying systems. There are questions of whether or not holiday code freezes are forcing organizations to hurry up and implement more changes than usual this time of year. Every organization handles its change windows differently but this could certainly be exacerbating the issue.

Optimistically I believe that the number of vulnerabilities that exist and are waiting to be discovered is finite, so (hopefully) we just have to wait for all the major organizations to shake out a few years of technical debt and we will return to relative normalcy. The “Patchpocalypse” to address the “Vulnpocalypse,” if you will.”

Patrick Harr, Chief Executive Officer, DataVisor:

“What stands out to me with these outages is the breadth. Banks, major technology companies, and government systems all experienced disruptions. Separate organizations can face the same pressure at the same time. When AI uncovers vulnerabilities faster than teams can test the fixes, systems that run reliably on an ordinary day can struggle with the pace of change. That’s a form of fragility we need to take seriously.

AI can find a flaw in minutes. It can take much longer to understand what happens to every service that depends on the software you’re about to change. Imagine a bank patches its sign-in system. The fix can close the security hole and pass its own tests, yet customers could still lose access if the mobile app can no longer authenticate them. That’s why I’d prioritize testing the whole customer journey and rolling the fix out in stages. Whether a particular organization missed a test is a question for its incident review.

Security teams need to change their monitoring, incident response and recovery plans. First, monitor what customers can actually do. Again, taking banks as an example, can customers sign in, see a balance, and complete a payment? Give the incident lead the authority to pause a rollout when those checks fail and rehearse how to restore service while keeping the vulnerable component isolated. Second, look across accounts for signs of coordination. One new account, one login, or one small transfer can look perfectly ordinary. When dozens of apparently unrelated accounts start moving money in the same sequence toward the same destinations, the pattern tells a different story.”

Chris Hughes, Vice President of Security Strategy, Noma Security:

“Artificial Intelligence (AI) has industrialized vulnerability discovery, organizations are racing to fix what’s being found, and the patching itself can disrupt operations. That’s the availability piece of the CIA triad. We’ve seen industrialized vulnerability discovery all year, in programs like Patch the Planet and Project Glasswing and in a steady stream of findings from frontier labs and researchers. Vulnerability trackers reported the 2026 CVE count passing 100,000 earlier this month, more than double last year and a record. That volume is outpacing the ability to patch and test safely, and when teams compress testing windows to keep up, they raise the risk of exactly these kinds of outages. The work that matters is risk-based prioritization around exploitability, known exploited vulnerabilities, and reachability, along with better testing, rollback plans, and resilience.

So, what technical indicators would support an AI-driven, coordinated attack as an explanation for these incidents? You’d want indicators that actually tie the incidents together: shared tooling or TTPs, artifacts consistent with autonomous offensive frameworks, and timing or targeting that can’t be explained by a common vendor, a maintenance window, or a shared patch cycle. Based on what’s public, we can’t say that’s what happened, and many of these disruptions have been described as planned maintenance or the result of accelerated patching. The coincidences are hard to ignore, which is why the speculation has taken off, but the better-supported story right now is accumulated technical debt and rushed remediation under the pressure of industrialized discovery. Organizations should still prepare for agentic offensive frameworks, even if this cluster isn’t ultimately attributed to them.”

Alex Orleans, Head of Threat Intelligence, Sublime Security:

“What I feel AI is more likely contributing to these outages is not patching speed or the discovery of vulnerabilities but code of dubious quality. As more and more development and engineering is devolved to vibe-coding, the internals of the resultant code are less and less understood by both the organizations deploying it and the engineers who maintain it. This makes crashes and outages more likely.

I would also say that across the outages we’ve seen, there hasn’t been a firmly identified shared type of cause (AI related or otherwise). Each outage should be taken on its own technical merits, not viewed as if there is some sort of unidentified overlap based on coincidental timing.”

Gerasimos Marketos, Chief Product Officer, Hack The Box:

“These incidents make security teams realize that attackers may use the same AI (used to discover the vulnerability x) to turn a published patch into a working exploit within hours. The answer is to give defenders and red teamers the same advantage.

A red team working with AI can show within hours which of a thousand flaws are actually exploitable in your environment, so only those few get the emergency treatment. A SOC working with AI can put detection and containment around a new flaw the same day, which buys the time to roll the patch out safely.”

Patrick Quirk, President, Opengear: 

“We don’t have specifics on these incidents, and outages have many causes, so I’ll speak to the pattern. AI is accelerating how quickly vulnerabilities are found, which means patches are arriving faster, and that speed is an advantage only when every rollout is tested in a staging environment first. Skip that step or miss a key dependency and the fix becomes the failure.

Speed doesn’t change the question that matters on the day something breaks: can your team get back into the affected systems and start recovery? That’s where many organizations fall short. They manage and troubleshoot their infrastructure over the same production network that just went down, so when it fails, IT and security lose access to the very systems they need to restore. In a cyberattack that dependency is dangerous, because visibility and control are what contain the threat.

Design your infrastructure assuming it will fail and build an independent management path that stays reachable when the primary network does not. That gives your team a way to diagnose the problem, isolate affected systems, and begin recovery without relying on the infrastructure that failed.”

Crystal Morin, Senior Cybersecurity Strategist, Sysdig:

“We swung the pendulum too far one way. For years, the security mantra was “patch faster.” Don’t get me wrong, that’s still the right approach. But many orgs have taken it to an extreme where the answer to every finding is “ship the fix, now, everywhere.” Pair that with a software delivery culture that already moves at breakneck speed, plus an AI-driven firehose of findings, and you get change volume that surpasses their ability (or desire?) to test and stage before deploying.

System fragility is the ultimate issue here. Systems are toppling over when you try to fix them because they were already broken. It’s like pulling the final brick in Jenga. These are architecture and infrastructure debts coming back to haunt us. AI-speed vulnerability discovery just delivered the bill to our doorstep.

Patching is prevention, and prevention is necessary, but it has a limit. What we’re seeing is what happens when you blow past that limit. So to bridge that gap while you fix your architecture, you need to turn to detection, which doesn’t have those limits.”

Originally published on SecurityWeek