Microsoft shipped fixes for 138 vulnerabilities this week — 30 of them rated Critical — and buried in the release notes was something quietly significant: 16 of those flaws were discovered not by human researchers, but by an internal AI system called MDASH.

MDASH stands for Multi-model Agentic Scanning Harness, which is a mouthful, but the concept is straightforward. It's a collection of specialised AI agents — each tuned for different vulnerability classes — that scan Microsoft's own codebase continuously and hand off findings for remediation. Sixteen CVEs in a single Patch Tuesday cycle isn't a rounding error. It's a signal.

And Microsoft isn't alone. Palo Alto Networks disclosed this week that its own internal AI tool, dubbed Mythos, has been finding dozens of vulnerabilities in its products. Meanwhile, the UK's AI Security Institute published an evaluation showing that OpenAI's GPT-5.5 — a generally available model — performs comparably to Mythos at vulnerability discovery. OpenAI has since launched a purpose-built security research product called Daybreak, positioning it as a more accessible alternative to Anthropic's tightly restricted Mythos offering.

What this means in practice

We are entering a phase where AI is measurably accelerating the rate at which software vendors find and fix their own bugs. The Record notes that Microsoft has already patched more than 500 vulnerabilities in the first five months of 2026, and is on pace to shatter its annual record. That sounds alarming until you consider the alternative: those bugs existing silently, waiting for someone else to find them first.

The wrinkle, of course, is that "someone else" has the same tools. The same frontier models being used defensively are available — or will be — to attackers. GPT-5.5 is already generally available. Daybreak is gated, but gating erodes over time. The asymmetry that defenders have historically needed — more resources, more time, more expertise — is being compressed from both sides.

What to watch

This Patch Tuesday had no confirmed zero-days, which is either reassuring or suspicious depending on your disposition. The two standout vulnerabilities to action immediately: a critical zero-click flaw in Outlook (CVE-2026-40361) that security researchers are already comparing to the 2015 "BadWinmail" bug — an enterprise-killer that spread malware via email without the recipient clicking anything — and separate patches from Fortinet and Ivanti for critical remote code execution flaws in products that are widely deployed across Australian government and enterprise networks. ACSC guidance on Ivanti vulnerabilities has historically followed within days of major disclosures; expect an advisory shortly.

The broader story is one of tempo. AI isn't replacing security teams — it's changing the rhythm of the work. Patch cycles that used to feel monthly are becoming continuous. The organisations that haven't automated their patching pipelines are already falling behind, and the gap is only going to widen.

For defenders right now: prioritise the Outlook zero-click, check your Fortinet and Ivanti versions, and treat this month's volume as the new baseline — not an anomaly.