Credit: bluestork / Shutterstock Microsoft’s September Patch Tuesday fixes 974 security flaws across its products, according to the company’s release notes. Attackers were already exploiting two of them before the patches landed on Tuesday. Both zero-days let an attacker who already has a foothold on a Windows machine gain higher privileges. One sits in the Windows Update Stack (CVE-2026-81963), the other in the Advanced Local Procedure Call component (CVE-2026-85880).
There is no public detail yet on who is exploiting them or how widely, as Dan Goodin reported for Ars Technica. Dustin Childs of the Zero Day Initiative, who reviews every monthly release, called it a new record. Childs doubts attackers have hijacked the update mechanism itself. His more likely reading is that attackers pair the Update Stack bug with a code execution flaw.
That combination can spread malware or ransomware. His advice on both is to patch quickly. A record that keeps being broken Counting is never exact. Microsoft’s notes list 974 of its own CVEs, the standard identifiers for disclosed flaws, and republish 25 more from other vendors.
Childs counts 972 new ones, or 997 if you add fixes for Chromium, the Google engine inside Edge. More than 110 are rated critical, Microsoft’s highest severity level. The totals have climbed fast. Microsoft fixed roughly 570 flaws in July’s record release and about 620 in August.
By Ars Technica’s count, it has now fixed 2,760 this year, more than double last year’s total. Childs opened his review with a phrase from his military days: “embrace the suck.” He congratulated the “security gnomes at Microsoft for being able to patch bugs at this rate”. Then he warned that “AI-assisted vulnerability discovery shows no signs of slowing down.” What to patch first Beyond the zero-days, Childs singled out a flaw in Exchange Server (CVE-2026-55007). An attacker with no login could run code on a mail server just by sending an email with a rigged Visio attachment.
The code runs when the server processes the message, so nobody has to open anything. Microsoft rates the exploit as unreliable. Childs points out that an attacker only needs it to work once. Outlook has a similar problem on the desktop.
An attacker can trigger CVE-2026-78509, rated 9.8 out of 10, through the Preview Pane. The victim only has to preview a message. Remote Desktop Services carries another 9.8 (CVE-2026-69525). A flaw in Microsoft Authenticator for Android (CVE-2026-80097) needs more work.
A malicious app on the phone has to trick the user into completing a sign-in. If it does, the attacker walks away with the user’s tokens. Childs called it the “worst type of privilege escalation” because it abuses the authentication system itself. The number network teams should care about most is 20.
That is how many bugs Childs classed as wormable. They need no login and no user action, so they can spread from machine to machine on their own. They sit in core services such as DHCP, DNS, Netlogon, Message Queuing and SMB. On domain controllers, a Kerberos flaw (CVE-2026-69676) would let any domain user run code with one crafted request.
Microsoft rates it as more likely to be exploited. Fuzzers, stamps and a Copilot bug Some components read like one tool pointed at one target. Windows DHCP Server alone has 36 fixes. Childs suspects someone aimed a fuzzer at the way it parses network packets, probably with AI help.
A fuzzer feeds software malformed input until it breaks. The Windows Biometric Service has 64 fixes, and 56 are the same kind of heap overflow: “one bug class, stamped 56 times,” in his words. AI also shows up on the other side of the ledger. Among more than 60 SQL Server fixes is CVE-2026-65669.
It works by persuading a user to submit crafted instructions to SQL Copilot, the AI assistant in SQL Server Management Studio. If that succeeds, the attacker reaches the database with the user’s permissions. The tools now used to find flaws have become a source of them too. Why the flood is happening The surge is partly deliberate.
Two weeks ago OpenAI, Anthropic, Amazon Web Services, Google, Microsoft and 100 other companies and organisations published an open letter. It warned that the window to fix flaws is closing before AI-enabled attackers exploit them first. Google and other vendors have also posted record fix counts in recent months, Ars Technica reported. Not everyone takes the AI story at face value.
Critics question what AI-assisted bug hunting costs and how many false positives it throws up. They also question the motives of AI companies that need a return on the billions spent building the models doing the hunting. The junk is real: in August, TNW reported on 55 vulnerability reports posted in days, 54 of them invented by AI. The strongest results cut the other way.
Mozilla has said Anthropic’s Mythos helped it find 271 Firefox bugs, with almost no false positives. What has not followed is a wave of attacks. Childs noted that “we have not seen a correlating spike in active exploits”, at least not yet. That fits a pattern TNW reported in July: AI is finding far more flaws, but almost none get exploited.
Europe’s clock starts tomorrow The timing matters in Europe. From 11 September, the EU’s Cyber Resilience Act requires makers of software and connected products to report actively exploited vulnerabilities. They must send an early warning within 24 hours of becoming aware, a fuller notice within 72 hours and a final report within 14 days of a fix. Actively exploited flaws like this month’s two zero-days are the category the rule targets.
Microsoft’s next scheduled release, on 13 October, will be its first full cycle under it. Also tagged with
