Microsoft Fixes a Record 1,413 Vulnerabilities in October, and Only One Is Known to Be Under Attack

Microsoft’s October 2026 Patch Tuesday fixed a record 1,413 CVEs, roughly 45 percent more than the September release we covered as the largest on record. Only one of them is flagged as exploited: CVE-2026-96951, a Windows Shell flaw that lets an attacker slip past Mark of the Web protections. CISA added it to its Known Exploited Vulnerabilities (KEV) catalog on October 13 and gave federal agencies until Friday, October 16 to patch.

Counts differ by a few depending on the rules. Microsoft’s release notes say 1,413 in their heading and 1,414 in the total row of their table, BleepingComputer counted 1,411, and the Zero Day Initiative (ZDI) counted 1,415. The release spans 176 distinct updates, 39 of them for Windows.

The one flaw under attack

Microsoft describes CVE-2026-96951 as “improper neutralization of special elements used in a command (‘command injection’) in Windows Shell” that “allows an unauthorized attacker to bypass a security feature over a network.” The feature is Mark of the Web (MotW), the tag Windows attaches to files that came from the internet so that, as SecurityWeek explains, Defender SmartScreen and Office Protected View can restrict automatic execution. Microsoft’s FAQ says an attacker “could evade Mark of the Web (MOTW) defenses.” Exploitation still needs a person: the attacker “would have to convince a user to take action,” through a link or an attachment sent by email or instant message. Microsoft scores the flaw 8.8 under CVSS 3.1, and NVD had not finished its own analysis when we checked.

We found no public account of the attacks; SecurityWeek notes that “no information has been shared about the zero-day attacks.” Microsoft’s feed lists four credit lines for the bug: “yhw,” “Benoit Sevens using Claude with Anthropic,” and two researchers “with Fujitsu Limited.” ZDI puts the bug in context. Going back to 2024, attackers have used several methods to avoid MotW protections, and “this time, it’s Shell instead of SmartScreen, but the result is the same.”

The fix ships in the October Windows updates for Windows 10, Windows 11 and Windows Server 2012 R2 through 2025, among them KB5127648 for Windows 11 24H2, 25H2 and 26H2 and KB5126657 for Windows 10 21H2 and 22H2. CISA’s entry, titled “Microsoft Windows Command Injection Vulnerability,” lists ransomware use as unknown and carries the forensic-triage flag. CISA’s implementation guidance says the requirement of BOD 26-04 “is that an adequate forensic triage analysis is performed,” and describes forensic triage as the way to assess “whether the systems or network infrastructure have been impacted or compromised.”

Two bugs were public before the patch, and neither is exploited

CVE-2026-108398 is a .NET Framework flaw that can bypass Windows Defender Application Control (WDAC). Microsoft says “an authenticated attacker could place a specially crafted .NET assembly on an affected system and register it for loading by a protected process,” scores it 8.8 and marks exploitation as “Exploitation Less Likely.” BleepingComputer credits James Forshaw of Google Project Zero, and Microsoft’s feed carries the same credit line. CVE-2026-92707 is a denial-of-service bug in the MessagePack-CSharp library, scored 5.3, whose record names GitHub rather than Microsoft as the CVE numbering authority. ZDI says it “was released by a third party and is now included in Microsoft’s release.”

Volume is up fourfold, exploitation is not

ZDI’s Dustin Childs counted “1,415 new CVEs from Microsoft,” which “puts this month at a greater volume than any previous YEAR from Microsoft.” SecurityWeek says the same about 2025 (“this month alone, Microsoft patched more vulnerabilities than it did in all of 2025”) and adds that the surge, “largely driven by AI,” has pushed 2026’s total past 4,000.

Microsoft’s public feed lets us check the trend directly. Counting Microsoft-assigned CVEs first published on each Patch Tuesday, January to October 2026 adds up to 4,081, against 1,008 for the same months of 2025, about four times as many. The number flagged as exploited did not follow: 16 against 20.

ZDI’s count of zero-days points the same way: 15 this year against 17 in the same months of 2023, a year in which Microsoft shipped 914 CVEs in total. CISA’s KEV catalog, a different measure, lists 40 Microsoft additions so far in 2026 against 39 in all of 2025 (our count of CISA’s JSON), which is roughly level, not fourfold.

Severity did not scale either: BleepingComputer counts 70 Critical fixes this month, 57 of them remote code execution, against 113 in September by our count of the feed. Elevation of privilege is the largest category at 677 of BleepingComputer’s 1,411. ZDI’s conclusion: “What matters is that discovery is scaling faster than exploitation, so teams that triage by sheer CVE count instead of exploitability are burning cycles on the wrong bugs.”

Microsoft’s answer is a grouping idea, not a new score

A same-day post from Microsoft proposes “risk cohorts,” defined as “a group of CVEs that require a similar customer assessment.” It lists 15 illustrative cohorts whose counts add up to 1,118 CVEs, the largest being Windows filesystems, storage and backup (136), Office document and mail clients (135) and Windows graphics, media and content processing (131). Microsoft stresses that the groupings “are not official Microsoft vulnerability categories, do not change Microsoft update guidance, and do not replace the assessment of individual CVEs,” and that the Security Update Guide does not currently display them. ZDI welcomed the idea while noting, “Back in my day, we called these security bulletins.” The post’s own figures drift: its table gives 135 CVEs for the Office cohort while the worked example cites 119, and 65 remote-access CVEs in the table become 64 in the example. It does not explain the difference.

What the credit lines say about AI

BleepingComputer says Microsoft “continues to use artificial intelligence platforms to find and remediate security vulnerabilities in its products,” and ZDI answers its own question, “Is AI to blame?”, with “Partly.” Microsoft’s feed offers one first-hand check, because each CVE carries credit lines that name who reported it, and some name the tools.

We counted credit lines that name an AI model, an AI lab, or an AI security firm or tool. They appear on 118 of October’s 1,409 Microsoft-assigned CVEs (8.4 percent), against 94 of 973 in September (9.7 percent), 41 of 571 in July (7.2 percent) and 4 of 163 in April. One name accounts for most of them: Diffract, whose site calls it an AI vulnerability research agent, appears in the credit lines of 78 of the 118. In another 23, the credited researcher says they used GLM from z.ai. The rest are XBREACH.AI (6), Theori’s Xint (5), Armadin (3), minsight-ai.com (1), Microsoft’s own MDASH agentic scanning system (1), and the Claude credit on the exploited zero-day (1).

These are credit lines that say so, which makes them a floor, and they cannot tell us how much of the surge AI explains. The share has sat between 7 and 10 percent since July even as the monthly count rose from 571 to 1,409. In October, 1,291 CVEs carry no AI-named credit and 85 carry no credit line at all. Microsoft links the two trends itself: “As AI-assisted research contributes to vulnerability discovery, this kind of approach may become more useful.” The receiving side is already straining elsewhere. Google’s open-source bounty stopped accepting product vulnerability reports this month, citing automated submissions, and Anthropic’s OSS Scanner emails maintainers bug reports that no human has reviewed.

What to do with 1,413 CVEs

Patch CVE-2026-96951 first. The KEV deadline is October 16 for federal agencies, and the exposure is the same for everyone else. After that, ZDI’s rule of thumb is: “Patch what’s proven, prioritized, and public, and treat everything else as this month’s new normal.”

SharePoint administrators should read up on CVE-2026-102866, which ZDI calls the only Critical-rated SharePoint CVE this month (CVSS 8.1) and the only one an unauthenticated attacker can exploit; it is an improper cryptographic signature verification bug that needs the attacker to have obtained a valid proof token for the target first. Our earlier SharePoint story covers an exploited SharePoint flaw from last month.

Machines running Windows 11 24H2 Home and Pro or Windows 10 Enterprise LTSB 2016 are out of updates after this month, because Microsoft says “the October 2026 monthly security update is the last update available for these editions,” so they need to move to a supported edition.

How we counted: we downloaded Microsoft’s public CVRF documents (for example the October 2026 document) on October 14 and counted CVEs with Microsoft as the numbering authority whose first revision falls on that month’s Patch Tuesday. For October that gives 1,409, within a few of Microsoft’s 1,413 and BleepingComputer’s 1,411. Credit lines were matched against names we checked by hand, and the feed is revised over time, so later counts can differ.