Dear Microsoft,
Yesterday you shipped the largest Patch Tuesday in history, 206 vulnerabilities in a single month, the biggest release since the program started in 2003. 33 rated Critical, 28 of those RCE. Multiple zero-days, some publicly disclosed or actively exploited before a fix even existed.
And here’s the thing I don’t think 206 is big enough. With the AI compute you’re sitting on, you should be out-finding the entire external research community combined for the last 23 years. A record Patch Tuesday driven by your own discoveries — your code, your fuzzers, your models, your fixes — would be a good thing. I’d happily patch more if it meant you found the bugs before the attackers did. Instead, 206 is what it looks like when outside researchers set the pace and you keep up. For a company with your resources, keeping up is not the bar.
Here’s the part I can’t get past. You have more AI compute than almost any organization on Earth. You’ve spent two years telling us Copilot will transform everything. And you’ve already proven it works for security — your own researchers used Security Copilot to find 20 previously unknown vulnerabilities in third-party bootloader code, including integer overflows that could bypass Secure Boot. You said it saved the team about a week of manual analysis.
That was someone else’s code…
So my question is simple: are you doing this against Windows, Exchange, SharePoint, and Entra at scale? From the outside, the evidence says no. Look at where this month’s headline zero-days came from — outside researchers, an offensive security firm, and in-the-wild exploitation. Not from you. If AI-assisted discovery works well enough for a victory-lap blog post about somebody else’s code, it works well enough to hammer on HTTP.sys, the Print Spooler, CTFMON, and every parser in the Windows kernel before the people writing exploit PoCs get there. Yesterday’s record patch load wasn’t exotic new bug classes. It was integer overflows, race conditions, and link-following privilege escalations — the exact patterns you already demonstrated AI is good at finding.
And here’s my other question, the one that really keeps me up at night: of the CVEs and zero-days being reported by external parties, how many are already sitting in your queue? How many did you already know about? We have no way to verify, because you publish nothing about your backlog — no queue size, no report ages, no time-to-fix. What we do see is researchers claiming they reported bugs through MSRC, got dismissed or cut off, and went public out of frustration — and the long-running pattern of reports labeled “no issue” that quietly get fixed months later without a CVE or an acknowledgment. When a “zero-day” drops publicly, we have no idea if it blindsided you or if it had been aging in your triage queue for a year. Honestly, I’m not sure which answer is worse.
The Secure Future Initiative promised security comes first, before features. Meanwhile, the visible AI investment goes into pinning Copilot into every product surface we then have to govern, while the patch counts go up — not down. That’s backwards. Every dollar of compute spent making Copilot harder to disable is a dollar not spent finding the next Exchange zero-day before an attacker does.
To be fair, discovery is the easy half. Finding 206 more bugs a month doesn’t help anyone if they sit in a fix queue. So here is the actual ask:
- Run AI-assisted vulnerability discovery against your own products, continuously. Windows, Exchange, SharePoint, Entra — the code we all depend on. Not a one-off research project. A standing program.
- Prioritize the critical path. Authentication, the kernel, and anything parsing remote input. That’s where the exploited zero-days keep landing.
- Ship the fixes before disclosure, not after exploitation. You have the compute, and you have codebase access no external researcher will ever get.
- Publish your queue. Open external reports, median report age, and time-to-fix. Right now the only visibility we get into your backlog is when a researcher gets frustrated enough to start dropping exploits publicly.
- Publish the numbers. How many vulnerabilities your own AI found in your own products, and how many shipped fixed before anyone outside ever knew. That’s a Secure Future Initiative metric I’d actually believe.
Until then, every record-breaking Patch Tuesday is a reminder that the AI revolution apparently applies to everything except the code the rest of us have to patch.
Sincerely, One of the admins patching your products this week
Tags: microsoft, patch tuesday, AI, security copilot, SFI, windows