Microsoft's Patch Tuesday Nearly Tripled to a Record 570 Flaws. Its Own AI Is Why.

TL;DR

Microsoft just shipped the largest Patch Tuesday in its history: 570 vulnerabilities fixed in a single day, nearly triple June's own record of about 206. Three were zero-days, and two were already being exploited in the wild. The reason for the surge is not a bad month. It is Microsoft turning an AI vulnerability-discovery system loose on its own Windows codebase, and the company is openly telling you this is the new normal. If you ship or defend software, the patch treadmill just changed speed.

What actually shipped

On July 15, Microsoft's monthly update closed 570 security flaws spanning Windows, Office, Azure, SQL Server, Exchange, SharePoint, Visual Studio, and a long tail of other products. Krebs on Security and BleepingComputer both put the count at 570. That is nearly three times June's 206, which had itself stood as the all-time high since the Patch Tuesday program began in 2003. Two consecutive record months, and the second one lapped the first.

Of the 570, roughly 59 carry a Critical rating, and 48 of those are remote code execution: the class where an attacker runs code on your box with little or no interaction. The rest skew heavily toward privilege escalation, which we will get to.

The reason is Microsoft's own AI

Microsoft is not blaming a rough quarter. It is taking credit. The company says it "has begun to use an AI-powered vulnerability discovery system to identify more security flaws across its Windows codebase before attackers can exploit them," and that it expects more Windows security updates from AI-discovered flaws going forward. Executive Vice President Pavan Davuluri put it plainly:

The pace of vulnerability discovery is changing with advances in AI making it possible to find more issues, faster, across more code, with new mechanisms that can accelerate both discovery and analysis.

Here is the part worth sitting with. A bigger patch count does not mean Windows suddenly got more broken. It means Microsoft is finally looking at more of it. Think of it like handing every code auditor a metal detector: you do not plant more landmines, you just start finding the ones that were quietly there the whole time. In technical terms, AI-assisted analysis widens the fraction of code paths actually inspected per release, so the count of discovered vulnerabilities climbs even if the true defect density is flat or falling. The flaws were always there. The floodlight is new.

The three zero-days you patch first

Records are fun trivia. The zero-days are the part that ruins your afternoon. Three shipped this month, and two are under active attack:

• CVE-2026-56155 (Active Directory Federation Services): an elevation-of-privilege flaw already exploited in the wild, letting an authorized attacker climb privileges locally.
• CVE-2026-56164 (SharePoint Server): missing authentication on a critical function lets an unauthorized attacker elevate over the network. Also exploited.
• CVE-2026-50661 (Windows BitLocker): a security-feature bypass that defeats device encryption. Publicly disclosed, not yet seen exploited, which is not a reason to sleep on it.

And one wry footnote from Krebs: among the criticals is CVE-2026-48561, a 9.6-severity remote code execution in Microsoft Copilot. The AI assistant that helps you write code needed a fairly serious patch of its own. Everyone is having a moment.

Why builders should care

This is not a Windows-admin story you can skim past. It is a preview of the whole industry's patch cadence. Trey Ford of Bugcrowd framed the shift bluntly:

AI has collapsed the cost of finding vulnerabilities, and this increase in volume is a new floor, not the ceiling.

Translate that into your week. Your patch window shrinks, because the gap between "flaw found" and "flaw fixed" is now gated by AI, not by human review time. Your test and regression matrix grows, because 570 changes at once is a lot of surface to validate before you roll them. And the dominant class this month is elevation of privilege (254 of 570), which is the quiet, boring category that turns a foothold into a full compromise. "We will patch it next cycle" gets more expensive every month it survives.

The other shoe: every large-codebase vendor is about to run this exact playbook against its own source. Expect the same firehose from the other big software shops and the major Linux distributions over the coming months. Patch Tuesday is quietly becoming Patch Firehose, and your automation had better be ready to drink from it.

The uncomfortable symmetry

The same class of tooling that let Microsoft surface these flaws first is the class attackers are pointing the other way. For now, the defender is winning the race: two of this month's three zero-days were caught and patched, and the whole point of the AI system is to reach the bug before someone else does. But "collapsed the cost of finding vulnerabilities" is a neutral sentence. It helps whoever runs the scanner. This month it was the vendor. The margin of that race is the thing to actually watch.

Key Takeaways

• Record scale: 570 vulnerabilities fixed in one Patch Tuesday, nearly triple June's prior record of about 206.
• Three zero-days: AD FS and SharePoint elevation-of-privilege flaws are already exploited; a BitLocker bypass was publicly disclosed.
• AI is the cause, by Microsoft's own account: an internal AI vulnerability-discovery system is surfacing more flaws, and the company says to expect more.
• Privilege escalation dominates: 254 of the 570 flaws are elevation of privilege, the class that turns a foothold into a takeover.
• New floor, not ceiling: Bugcrowd's read is that AI-driven discovery volume only grows from here, and every big vendor is next.
• Do this now: prioritize the exploited zero-days, tighten your patch window, and assume the firehose is permanent.

Sources: Krebs on Security, BleepingComputer, Infosecurity Magazine, Microsoft Security Response Center