Analysis · Security & Fraud

AI Is Finding Software Flaws Faster Than Organizations Can Safely Fix Them

Microsoft's monthly security fixes rose from 113 in January to 1,407 in October, against a 2025 record of 167. Whether or not rushed patching caused this month's outages, AI-speed discovery is turning patch management into a capacity problem.

By The AI Enterprise Magazine Team7 min readAnalysis · Weekly briefing

Illustration of a shield covered in patches under a rising stack of alerts
Illustration of a shield covered in patches under a rising stack of alerts

Key takeaways

  • A ninefold jump. Tenable counted 113 Microsoft fixes in January 2026 and 1,407 in October, against a prior record of 167 in October 2025.
  • AI is a stated driver. Microsoft said its agentic scanning system would bring higher volumes of security updates.
  • Triage is the bottleneck. Anthropic says verifying, disclosing and patching is now the constraint.
  • Patch theory unproven. No company has confirmed that rushed patches caused this month’s outages.

Special Report · Outage Week

Microsoft's Monthly Security Fixes Have Grown More Than Tenfold in 2026

Microsoft fixed 113 vulnerabilities in January and 1,407 in October, against a previous single-month record of 167 in October 2025. Microsoft says its AI-driven scanning is one reason for the higher volume.

Microsoft's Monthly Security Fixes Have Grown More Than Tenfold in 2026113Jan54Feb83Mar163Apr118May198Jun569Jul398Aug964Sep1407Oct1093Nov9.7x from Jan
Number of CVEs addressed in each monthly Microsoft Patch Tuesday release in 2026, as counted by Tenable. Tenable omits a small number of third-party or externally assigned CVEs; other trackers' totals differ slightly. The previous record before 2026 was 167, in October 2025. Source: Tenable monthly Patch Tuesday analyses
View the data
CVEs fixed in Microsoft's Patch Tuesday release, 2026
Jan113
Feb54
Mar83
Apr163
May118
Jun198
Jul569
Aug398
Sep964
Oct1407
Nov1093

More in Data Viz

In October 2025, Microsoft set a record for the largest monthly security release in its history: 167 vulnerabilities fixed on a single Patch Tuesday. Twelve months later, that record looks like a figure from another era. In October 2026, Microsoft fixed 1,407 vulnerabilities in one release. In November it fixed 1,093. Those are the counts published by the security firm Tenable, which has tracked Microsoft's monthly releases for years, and they describe a change in the basic economics of software security that most organizations have not yet absorbed.

It is not known whether this flood of fixes had anything to do with the outages that struck the FAA, Microsoft, JPMorgan Chase, Apple and the NSA in mid-November. Every organization involved has given its own explanation, and none of those explanations mentions AI. But the theory that has gained the most traction among lawmakers, the one Senator Mark Warner put into words, holds that AI is finding vulnerabilities faster than organizations can safely fix them, and that the rush to patch is breaking things. Whether or not that theory explains this particular week, the numbers behind it are real, and they are growing.

How fast has the volume grown?

Tenable's monthly counts tell the story with unusual clarity. In the first half of 2026, Microsoft's releases looked broadly normal: 113 vulnerabilities in January, 54 in February, 83 in March, 163 in April and 118 in May. June set a new record at 198. Then the curve bent. July brought 569, which Tenable described as "the largest Patch Tuesday release ever," followed by 398 in August, 964 in September, 1,407 in October and 1,093 in November, according to Tenable. Counting methods differ slightly across trackers, and other firms put some months a little higher or lower, but no one disputes the shape.

Microsoft has been explicit about one reason. In its July release, Tenable noted, Microsoft said it was using a multi-model agentic scanning system, which it calls MDASH, to find vulnerabilities faster, and warned that "customers will see a higher volume of security updates included in each security release." The one vulnerability exploited before November's release, a spoofing flaw in Outlook, was credited in Microsoft's advisory to a researcher who used Anthropic's Claude, Tenable reported.

Where are all these vulnerabilities coming from?

Microsoft is not alone. In April, Anthropic began giving a group of roughly 50 partner organizations access to Claude Mythos, a model built to find software flaws, under a program called Project Glasswing. By late May, Anthropic said the partners had found more than 10,000 high- or critical-severity vulnerabilities, CyberScoop reported. Mozilla alone fixed 271 vulnerabilities in a single Firefox release. Help Net Security listed the early partners as including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia and Palo Alto Networks. Anthropic later expanded the program to about 150 more organizations in more than 15 countries.

The presence of Microsoft, Apple and JPMorganChase on that list has fueled speculation online, since all three suffered outages this month. There is no public evidence connecting their participation in Glasswing to their outages, and the companies have attributed their failures to a July update bug, an internal update and an on-premise software bug respectively. What the list does show is that the organizations with the most sophisticated security programs have been handed tools that surface more vulnerabilities than they have historically had to fix.

Anthropic itself named the consequence. In its expansion announcement, the company wrote that "the bottleneck in cybersecurity is now verifying, disclosing, and patching" the vulnerabilities these models find. In May it told CyberScoop that the bottleneck in fixing such bugs "is the human capacity to triage."

Why would more patches cause more outages?

Every patch is a change to a running system, and every change carries some risk of breaking something. Software teams manage that risk with testing, staged rollouts, maintenance windows and rollback plans, all of which take time and people. When the volume of required changes rises tenfold, one of two things happens. Either the organization accepts a longer delay before fixing known flaws, leaving them open to attackers who now have the same AI tools, or it compresses the testing and rollout process to keep up.

Banking executives have been describing that dilemma for months. In April, Moneywise, in an article syndicated by Yahoo Finance, reported that banks were worried about the "patch gap," the window between a vulnerability's discovery and the fix being applied, noting that banks can take days or weeks to test and install updates across their systems. The same report cited CrowdStrike's finding that the average time from an attacker's first access to malicious action had fallen to 29 minutes. Faster attackers and more vulnerabilities create pressure to patch faster, and patching faster means testing less.

That is the mechanism Warner described. "AI is making it possible to identify vulnerabilities," he said, according to Implicator, and companies are rushing to patch all of them at once. The Washington Sun reported that bank executives and NSA officials told Congress something similar, citing three anonymous sources and a congressional official who described an effort "to push through a lot of patches all at once." No company or agency has confirmed that account publicly.

How strong is the evidence for the patch theory this week?

It is suggestive rather than conclusive. The clearest on-record statement came from Director of National Intelligence Jay Clayton, who told Fox Business that what the country was seeing was "just an upgrade in patching and upgrade in systems," according to Fox News. Implicator reported that Clayton tied the NSA's network failure to a patch for PHP vulnerabilities disclosed in October, which it says were found with AI help, though the published excerpts of his interview do not include that detail.

Against that, Microsoft traced its failure to a bug introduced in July, not a rushed November fix, and the FAA's ground stop followed a router installation that had nothing to do with vulnerability patching. Christiaan Beek of Rapid7 told Implicator that nothing public shows the outages came from rushed security updates.

The honest summary is that the patch theory fits some of this week's events loosely, fits others poorly, and has not been confirmed for any of them. The underlying trend, however, does not depend on this week at all.

What does this mean for enterprise technology leaders?

The patch volume of 2026 is not a temporary spike. Anthropic has said it expects other AI developers to have models with similar capabilities within six to twelve months, which means attackers will have them too. Every enterprise running commercial software, open-source libraries or its own code should assume that the rate at which serious vulnerabilities are discovered in its estate will stay several times higher than it was in 2025.

That changes the planning problem in three ways. First, patch management can no longer be treated as routine operations staffed for steady-state volume. It is now a capacity problem, and organizations need to know how many changes per month their testing and deployment processes can absorb safely, and what happens when demand exceeds that number.

Second, prioritization has to get much sharper. Not every one of the 1,093 vulnerabilities Microsoft fixed in November applies to every organization, and not every applicable one is equally urgent. Teams that can quickly determine which flaws are reachable and exploitable in their own environment will be able to slow down where it is safe to do so and move fast only where it matters.

Third, the same AI capabilities that are creating the problem have to be part of the answer. If models can find vulnerabilities at machine speed, organizations need machine-speed help to triage them, test fixes and verify that a patch has not broken something else. Anthropic says its own security tooling has been used to patch more than 2,100 vulnerabilities in three weeks. The question for enterprise leaders is not whether to use AI in vulnerability management, but how to govern it so that automated fixes do not become a new source of failure.

What should leaders take away from this week?

The mid-November outages may turn out to have nothing to do with AI-driven patching. The investigations are not finished, and the explanations offered so far point to ordinary failures of change and capacity. But the conditions that would make such failures more frequent are already in place. Software is being found to be more flawed than anyone knew, at a pace that no human-staffed process was designed for, and every fix is another change to systems that are already hard to change safely.

Enterprise leaders who wait for proof that AI caused an outage before adapting their patching, testing and change processes are waiting for the wrong signal. The chart of Microsoft's monthly fixes is signal enough.

Frequently asked questions

How many vulnerabilities did Microsoft fix in October 2026?

1,407, according to Tenable, the largest Patch Tuesday release on record. November followed with 1,093.

Why are there so many more vulnerabilities in 2026?

AI models are finding flaws far faster. Microsoft has cited its own agentic scanning system, and Anthropic says Mythos partners found more than 10,000 high- or critical-severity flaws in about a month.

Did AI-driven patching cause the November 2026 outages?

It has not been shown. Some lawmakers and an anonymously sourced report point to rushed patching, but Microsoft, the FAA, Chase and Apple attributed their outages to other causes.

Filed under: Security & Fraud, AI Governance & Trust, Data & AI Infrastructure

More from AI Enterprise Magazine

Read the magazine