Skip to Content (Press Enter)
Is your Statamic site hiding the pain like Harold? Find out

Patchmageddon and why AI has everyone patching at once

AI is finding vulnerabilities faster than they can be fixed. Install security fixes quickly, and check your site still works with each one before it goes live.

by Stephen Meehan in Ai, Maintenance Services, News

A conveyor belt feeds plain boxes into a machine amidst a sea of discarded cardboard.

Patchmageddon is the name JPMorgan gave in July 2026 to a flood of security fixes it saw coming, as artificial intelligence (AI) found vulnerabilities faster than companies could fix them. In the week of 12 November 2026, it arrived.

Microsoft 365, US air traffic control, Chase, the National Security Agency (NSA) and Apple all went down within six days. None of them was hacked. Each outage followed a change the organisation made to its own systems, and at the NSA, the US government linked it to an urgent security fix.

I spent the evening of 18 November updating NGINX, the software that serves every site I look after, a few hours after its fix came out. The next afternoon, PHP released fixes for 17 more vulnerabilities.

What is Patchmageddon?

Patchmageddon is JPMorgan's word for what happens when AI finds security vulnerabilities faster than people can fix them. Michael Cembalest, who chairs market and investment strategy at J.P. Morgan Asset and Wealth Management, used it as the title of a July 2026 report.

Anthropic, the company behind the Claude AI models, shows the scale of it. Its disclosure dashboard tracks the vulnerabilities its AI has found in open-source software. By 2 November 2026, it had reported 5,087 of them to the people who maintain that software. To its knowledge, 629 had been fixed.

For every vulnerability fixed, about seven more were still waiting.

Microsoft shows the same pattern. Its monthly security update in November 2025 fixed 63 vulnerabilities. A year later, the November 2026 update fixed more than 1,000.

Twelve security incidents from autumn 2026

Between October and mid-November 2026, these twelve made the news. Some were attacks. Most of the biggest were not. I track the ones that touch Statamic sites on Statamic Watchtower.

Vulnerabilities found, and attackers using them

  • PHP2Shell, 21 October. PHP, the programming language behind around 70% of websites, fixed two vulnerabilities that could let a stranger run their own code on a server without logging in. Researchers found them with the help of AI, and one had been in PHP for 14 years. I patched every site I maintain the same day.

  • Supermarket freezers, from October. Attackers took control of internet-connected refrigeration systems in hundreds of grocery stores across France, Belgium, Ireland, Luxembourg and Sweden, spoiling stock and in some cases starting fires. At least one vulnerability they used had a published fix from 2025. On 5 November, US agencies told businesses to take these systems off the internet, install updates and change default passwords.

  • Microsoft's November update, 10 November. More than 1,000 fixes in one day, including one for an Outlook vulnerability attackers were already using.

  • Microsoft's AI agent toolkit, 17 November. Someone published two poisoned versions of Microsoft's agent-framework package to the Python Package Index (PyPI), where developers download Python code. Installing either one handed over the keys to the developer's Azure and GitHub accounts.

  • NGINX, 18 November. NGINX, the software in front of around 3 in 10 websites, fixed ten vulnerabilities. Researchers using AI found the worst, which had been there since 2015, and working attack code was public the same day. Every site on my maintenance plans was updated the same week.

  • PHP again, 19 November. PHP fixed 17 more vulnerabilities. The worst is in PHP-FPM, the part of PHP that runs a site's code behind NGINX.

  • Chase's suppliers, 19 November. JPMorgan Chase said attackers had broken into outside companies that hold its marketing data, and may have taken customer names and email addresses. Chase said its own services weren't affected.

Chase Verified organisation @Chase , view the post on X
We have learned that third parties who provide services to the firm were the target of a cyber attack. Based on our investigation so far, we believe the attackers accessed certain marketing data held by those vendors, which may include some customer names and email addresses. There is no disruption to our services. Where appropriate, we will contact affected customers directly and offer support, including complimentary credit monitoring. There is no connection to service disruption some customers experienced earlier this week.

Outages caused by updates, not attackers

  • Microsoft 365, 12 November. Outlook and other Microsoft 365 services were down for about a day. At the peak, more than 80% of the servers behind one of its networking services had stopped responding. Microsoft's early report points to a hidden defect from a July 2026 update, and Microsoft said it was not a cyberattack.

  • US flights, 13 November. The Federal Aviation Administration (FAA) stopped every departure in the US for over four hours. A contractor had been swapping old copper lines for fibre, and the new router at one site overloaded the air traffic control network.

  • Chase, 16 November. Customers of America's biggest bank had trouble moving money for much of the day. Chase blamed a software problem in its own systems, unrelated to the supplier breach three days later.

  • The NSA, 17 November. The NSA's classified network went down. The Director of National Intelligence, Jay Clayton, said patching after PHP2Shell "caused some of these systems to not come back up". The NSA called it scheduled routine maintenance.

  • Apple, 17 November. iCloud, Apple Pay and the App Store went down the same day. Apple blamed an internal software update and ruled out an attack.

Why are security fixes causing outages?

Security fixes cause outages when they go live faster than they can be tested. Every fix changes code, and a change can break something its author didn't expect.

Trail of Bits, a security research firm, checked its own work in October 2026. Of more than 130 fixes its engineers had written for other people's software, 21 had broken something else, about one in six. None of the 21 weakened security, but each one was a new bug someone had to find and fix.

The time allowed for fixing is shrinking too. Google's Project Zero team gives companies a deadline to fix a vulnerability before it publishes the details. In November 2026 it announced that the deadline will fall from 90 days to 60 in February 2027, and to 30 days in January 2028. Anthropic is cutting its own deadline for serious vulnerabilities from 90 days to 30, starting in January 2027.

More vulnerabilities and less time to fix them means more changes going live at once, with less testing behind each one. That is how a security fix takes down a network.

Why AI is a double-edged sword for security

AI is a double-edged sword for security because the tools that help defenders find vulnerabilities help attackers use them. PHP2Shell and the worst NGINX vulnerability sat unnoticed for years until researchers using AI found them. That's the good edge. Old bugs are being found and fixed at a pace no team of people could manage.

Attackers have the same tools. In October 2026, DeepSeek released a model anyone can download and run on their own hardware, and Artificial Analysis ranked it the strongest openly available model for cyber security work.

A published fix also shows attackers where the vulnerability was. As I wrote in why the patch gap matters more now, AI has made it quicker to work backwards from a fix to an attack. Every unpatched copy of the software is exposed from the moment the fix is public.

Jason Clinton, deputy chief information security officer at Anthropic, warned in August 2026 of a "vulnerability tsunami" with too little time to patch critical systems. In October he said it would get worse before it got better, over roughly the next 18 months.

Jason Clinton Verified account @JasonDClinton , view the post on X
as bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better. security leaders, never let a crisis go to waste: please go to your boards and ask for help right now. not for ai, for the basics. (we need to hold on for ~18 months.)

What Patchmageddon means for your website

For a website, Patchmageddon means doing two things at once. Install security fixes quickly, because each fix tells attackers where to look. Check the site still works with each one before it goes live, because a rushed fix can take a site down as surely as an attacker can. The people who maintain PHP, NGINX and Statamic test their own fixes before releasing them. What I test is whether your site, its Control Panel and the server still work once a fix is installed.

Most of November's big outages weren't attacks. They were changes that went live before anyone found what they broke. A website carries the same risk on a smaller scale. A PHP update that clashes with an old package, or a server change that stops the Control Panel loading, can take a site offline for hours.

Patching slower isn't the answer. Checking the site on a staging copy after every fix is.

On my Statamic maintenance plans, every update goes on a staging copy of the site first. I check the site and its Control Panel work there, then install it on the live server. Critical security fixes don't wait for the monthly update, which is why PHP2Shell and the NGINX fixes went on in the week they came out.

If someone else looks after your site, ask them two things. How quickly does a critical security fix go on, and where do they check your site still works before it reaches the live server?

19th November, 2026 by Stephen Meehan in Ai, Maintenance Services, News
.

Get a measurably better website

Your online presence matters, increase engagement, lower bounce rates, and improve conversions.
Statamic Websites