By ByteBot
Microsoft’s September 2026 Patch Tuesday arrived with 974 security fixes — the largest monthly patch release in the company’s history. That’s 71% more than July’s record of 570, and more than double everything Microsoft patched in the entire year of 2020. The cause isn’t a sudden explosion in bad code. It’s MDASH: Microsoft’s multi-model agentic AI scanning system, which finds vulnerabilities at a speed human researchers never could. The problem is that discovery being fast doesn’t make deployment any easier.
Two of those 974 bugs are already being exploited in the wild. They need to go to the top of your list before anything else.
Two Zero-Days, Both Being Exploited Now
CVE-2026-81963 lives in the Windows Update Stack — the system responsible for keeping your machines patched. The irony is sharp: a flaw in the update mechanism means attackers can target fully patched systems by exploiting the very process designed to protect them. It’s a local privilege escalation (CVSS 7.8), meaning an attacker who has already gained a foothold via phishing or another remote code execution can use this to escalate to SYSTEM. Affected platforms: Windows 11 (23H2 through 26H1) and Windows Server 2025. For a detailed CVE-2026-81963 analysis, the attack chain is well-documented.
CVE-2026-85880 is a heap buffer overflow in Windows Advanced Local Procedure Call (ALPC), allowing AppContainer sandbox escape to SYSTEM privileges. It hits older systems: Windows 10 (1607 through 22H2) and Windows Server 2012 through 2019. Both flaws landed on CISA’s Known Exploited Vulnerabilities catalog on September 8, with a September 22 remediation deadline for US federal agencies.
Neither requires remote access to trigger — attackers need initial local access first, making these second-stage weapons in a longer attack chain. Patch them anyway. Working exploit code is out in the wild for both.
The Other Patches That Cannot Wait
After the zero-days, the next priority is CVE-2026-69829: a Windows Shell remote code execution with a CVSS score of 9.8. No user interaction required, no elevated privileges needed. This one should be on every Windows endpoint before the zero-days are even fully rolled out to legacy systems.
Infrastructure teams should also be aware that this release patches 17+ unauthenticated network-reachable remote code execution vulnerabilities across DNS, DHCP, Netlogon, Kerberos, and SSTP VPN. If these services face the internet directly, they are your perimeter exposure right now. Additionally, Windows Biometric Service received 64 fixes this month — the most for any single component in 2026 — which matters if you’re deploying Windows Hello or biometric-gated access controls. The complete CrowdStrike breakdown covers all 17 infrastructure CVEs with prioritization guidance.
Why AI Is Responsible for This Surge
Microsoft’s Autonomous Code Security team built MDASH (Multi-model Agentic Scanning Harness) in early 2026. The system runs more than 100 specialized AI agents against Windows binaries. Candidate vulnerabilities get debated across multiple agent families to filter false positives, and a “prover” pipeline generates proof-of-concept triggers to confirm bugs before they ever reach a human engineer. In its initial run, MDASH found 16 Windows vulnerabilities — including four critical remote code executions — that manual review had missed.
That’s the story behind the record numbers. AI doesn’t just find bugs faster. It finds entire families of bugs in one scan. Where human researchers would identify one symlink traversal issue, MDASH finds 40 variants of the same underlying pattern across the codebase simultaneously. Microsoft acknowledged this directly in July 2026: “The pace of vulnerability discovery is changing with advances in AI making it possible to find more issues, faster, across more code.” This same pattern is playing out with other vendors — Adobe, Cisco, Google, and Oracle have all credited AI-assisted research for increased patch cadence in 2026. It’s also why the broader AI security threat environment is escalating, as covered in our post on Zero-Click RCE in AI Coding Agents.
The Patch Fatigue Problem Is Now Structural
AI removed the human rate-limiter on vulnerability discovery. It did not remove the human rate-limiter on patch deployment. Organizations that could reasonably keep pace with 100-200 patches per month now face nearly ten times that volume. Security researcher Jack Bicer put it plainly: “The challenge is not simply getting through the patch list. It is knowing what needs attention first.” Another researcher described the situation as AI-assisted discovery “creating larger haystacks, but not finding more needles.”
This is a structural problem, not a temporary spike. Microsoft expects volume to stay elevated. The only viable response is triage: exploit status first, CVSS score second, exposure third. Tools like Windows Autopatch and Azure Update Manager help automate deployment prioritization, but the risk decisions still require human judgment. Organizations that haven’t adopted risk-based patch management — rather than time-based monthly cycles — are already falling behind. This is also relevant to the supply chain security context we covered in our OpenAI breach analysis: attackers exploit the gap between discovery and deployment.
One More Deadline on Your Calendar
If you’re running .NET 8, add November 10, 2026 to your calendar. That’s end-of-support for .NET 8 alongside PowerShell 7.4. It’s not directly a Patch Tuesday issue, but this month’s release includes regression risks in printing, USB device handling, and Remote Desktop sessions — the systems you want stable before migrating runtimes. Start the .NET 9 or .NET 10 migration now while this environment is still predictable.
974 fixes. Two active exploits. One AI system that isn’t slowing down. The patch math only gets harder from here.
