Quick Answer: Why everything keeps breaking comes down to a patching squeeze. AI tools now find and exploit software flaws faster and more cheaply than before, so a large backlog of old bugs is surfacing at once. Companies have hours, not weeks, to respond. Rushed updates and AI-written code can also cause failures of their own.

Every few weeks, another app, bank, airline system, or smart appliance goes down. Microsoft Outlook had a bad day. Air traffic control systems had issues. Chase had a rough Monday. It feels like a coincidence, but the pattern is real. As CNET reported, the answer to why everything keeps breaking isn't one giant cyberattack. It's a set of pressures that reinforce each other, and AI sits at the center of most of them.

The vulnpocalypse is a backlog, not a single wave

The most useful number in the story is a count of disclosures. The National Vulnerability Database received more than 17,000 Common Vulnerabilities and Exposures (CVEs) in October 2026. In October 2025, the figure was under 5,000. By mid-November, it had already passed 10,000 for the month.

PeriodApproximate CVEs logged in the NVDWhat it signals
October 2025Under 5,000Baseline before AI-driven discovery accelerated
October 2026More than 17,000Sharp rise as AI models scanned old code
November 2026 (to date)More than 10,000Pace continues into the following month

Matt Wyckhouse, CEO of Finite State, put it plainly: AI is finding bugs in open-source software that sat unnoticed because nobody had the time to look. That's the "vulnpocalypse" in practice. These flaws aren't new. They're just being found in bulk.

Here's the counterintuitive part. Most people assume the danger comes from new, clever attacks. In many cases, the danger comes from old code that was always weak and nobody had reason to examine closely. AI turned a slow manual review into a cheap automated one.

Hacking got cheap enough for anyone

Until recently, breaking into a serious system took a practiced team, often backed by a government or criminal organization. CNET's sources described how that has changed. Things that once required a team with a clear goal can now be done by one person with an AI model and a purpose-built program around it.

John Strand, CEO of Black Hills Information Security, described the shift from the defender's side. AI agents now handle the "grinding grunt work" of offensive security, which means a skilled operator gets through the routine parts much faster. Strand's point was that the automation is helping the attacker's workflow, not replacing the attacker's judgment.

The tooling is also getting sharper. A "harness" wraps a general AI model with the tools and instructions needed for one job. CNET reported that the FBI issued a warning about a harness called Artex. CrowdStrike, according to Adam Meyers, traced a major hack of South Korean banks in September and October to a suspect who was likely a 26-year-old in China using that tool. The attacker reportedly asked the AI how to carry out the intrusion, then asked how to sell the stolen data on Telegram.

Meyers's summary is blunt: "Anybody out there can now be a nation-state-level hacker just by using the tool." Vendors restrict their most capable models, but the skills needed for many exploits are within reach of open-weight models that attackers can run privately.

Defenders have hours, not weeks

Before, a newly discovered flaw could take weeks to turn into a working attack. Now it can take days or hours. That compresses the entire response chain: detecting the issue, testing a fix, deploying it, and confirming nothing else broke.

Microsoft CEO Satya Nadella told the New York Times DealBook Summit that the old workflow no longer works. Previously, you might have had a week to test a critical patch before pushing it to production. Now, he said, "the patch needs to go instantly from being available to being patched in production because it's being exploited in the wild."

Defenders are not standing still. Programs such as Anthropic's Project Glasswing, OpenAI's Daybreak, and Google's Fairwind give banks, healthcare providers, and security teams early access to strong models, often with fewer restrictions on what those models can search for. The PHP vulnerability that prompted an urgent notice from the PHP Foundation came out of work through Patch the Planet, part of Daybreak.

If you run a site or manage any internet-connected system, here's what matters most right now:

  1. Inventory everything that's connected, including devices you forgot about.
  2. Mark which components are end-of-life or no longer maintained by their creators.
  3. Set up a staged rollout so you can test a patch on a small group before everyone gets it.
  4. Monitor vendor advisories for critical flaws so you can act within hours, not after a week-long review.
  5. Keep a documented rollback plan, because the fix can break things too.

Step five matters more than most guides admit.

The fix can break things too

The common advice is "patch immediately." That's usually right, but it skips a practical problem. Software updates are believed to be responsible for several of the recent high-profile outages, including Apple's problems earlier this week. The attack isn't always what takes a system down. Sometimes it's the emergency fix.

This is the trade-off nobody wants to hear. Waiting a week to test a patch may leave you exposed to an exploit that's already being used. Pushing a patch instantly risks a regression that takes your service offline. Neither choice is free.

The middle path is a staged rollout. Push to a small canary group first, watch error rates and logs, then expand. That approach still moves in hours, not weeks, and it limits the damage if the patch is flawed. The tradeoff is more operational effort, which is why small teams struggle with it.

AI-written code is adding new holes

The second AI effect is quieter. Models trained on enormous amounts of code are very good at producing software that works. They are not reliably good at producing software that's secure. Code that runs can still have holes, and there's a real chance no human reviewed it closely before release.

CNET reported that companies are shrinking engineering teams while leaning on AI to write more code. Microsoft's October announcement that it was offering buyouts to 60,000 staff became a talking point weeks later, when its Outlook service went down for a day. The article doesn't establish a causal link between the two, and you shouldn't assume one. But it does raise a fair question: might things break less often if more experienced people were maintaining the systems?

Ryan McCurdy, vice president of marketing at Liquibase, raised a related concern. Junior developers may be shipping code that once would have gone through senior review, because AI boosts productivity but not experience. The problem isn't only old vulnerabilities. It's new ones being written into fresh software and patches.

Here's a practical test. Before you merge AI-generated code, ask who reviewed it, which tests cover the risky paths, and whether a static analysis tool such as Semgrep or CodeQL ran against it. If the answer to all three is "nobody," you're shipping on hope.

Connected devices are the scariest edge

Websites get the headlines, but the widest exposure may be in connected devices. Refrigerators, smart home gadgets, and industrial equipment often run firmware that few people have ever examined. Its source code may not be public, and it may no longer receive updates.

An AI model can reverse-engineer that firmware and search for exploitable flaws, which removes much of the protection that obscurity once provided. Wyckhouse noted that these devices probably didn't need to be online at all, yet they are.

The stakes are physical. Employees at a supermarket in Sweden had to empty freezers after a cyberattack pushed the temperature to 50 degrees Celsius. Wyckhouse's warning sums it up: "Refrigeration has to work for our society to function, as it turns out." Infrastructure that once depended on isolation now depends on patch discipline.

Frequently Asked Questions

What is the vulnpocalypse?

The vulnpocalypse is a informal term for the surge in discovered software vulnerabilities driven by AI tools that can scan code faster and more cheaply than human reviewers. Reported CVEs jumped from under 5,000 in October 2025 to more than 17,000 in October 2026. Many of these flaws existed for years, but few people had time to find them.

Why does software keep breaking after updates?

Updates are often emergency fixes written under pressure, and they can introduce regressions. Apple's recent outages were among several linked to software updates. Staged rollouts, automated testing, and rollback plans reduce the risk without slowing security fixes to a crawl.

Why does everything keep breaking right now, and is AI the main cause?

AI isn't the sole cause, but it's a major accelerant. It makes exploits cheaper to build and vulnerabilities faster to find, and it speeds up code output that may not be reviewed well. The result is more urgent patches arriving at a time when defenders have less time to test them.

How can a small team protect a website from AI-driven attacks?

Start with an inventory of every connected component, then remove or isolate anything unmaintained. Subscribe to vendor security advisories, use automated dependency scanning, and deploy patches to a small canary group before a full rollout. Most small teams can't match attackers' speed, but consistent basics block many automated attacks.

Can AI fix these vulnerabilities too?

Yes, and defenders are already doing it. Programs like Project Glasswing and Daybreak give security teams early access to powerful models to find and patch flaws first. The catch is that defenders must be right every time, while attackers only need to succeed once.

Stop treating outages as isolated events

Here's the bottom line: why everything keeps breaking isn't a mystery of random bad luck. It's a patching problem colliding with cheap AI-driven discovery and rushed code. Start this week by listing every internet-connected system you run, then mark which ones you can't patch quickly. Pass this to someone wrestling with a backlog of security alerts and no clear plan for testing them.