Friday, Oct 9 | --:--
Back to home

MDASH: Microsoft’s Agentic Scanner Finds 16 Windows Vulns — Same Day as Patch Tuesday

Microsoft Security’s MDASH — 100+ specialized agents in a multi-model scanning harness — helped find 16 Windows vulnerabilities (4 critical RCEs) shipping in the May 12 Patch Tuesday wave, while topping CyberGym-style benches in limited private preview. Dual-use agent stacks now sit next to the patch calendar.

Times of AI Desk 5 min read Redmond, WA View as Markdown
Cover illustration for MDASH: Microsoft’s Agentic Scanner Finds 16 Windows Vulns — Same Day as Patch Tuesday

Agentic coding tools are dual-use: the same stack that ships features can hunt CVEs. Putting MDASH next to Patch Tuesday makes “AI found the bugs we fixed today” a hyperscaler security story — not a research demo.

Microsoft Security (May 12) published “Defense at AI speed,” introducing MDASH (multi-model agentic scanning harness) from the Autonomous Code Security team. Microsoft said MDASH helped researchers identify 16 new Windows vulnerabilities in networking and authentication — including four critical remote code execution flaws — that went out in the May 12 Patch Tuesday wave. The system orchestrates 100+ specialized AI agents across an ensemble of models; in use by Microsoft security engineering and a limited private preview for select customers.

What MDASH is

Claim (Microsoft primary) Detail
Architecture Harness of specialized agents (discovery, validation, exploit proof, remediation assist) — not a single model
Results at launch 16 Windows vulns; 4 critical RCE-class; networking/authentication focus
Benchmark Topped a leading CyberGym-style industry leaderboard score at announcement
Status Internal production use; small customer private preview

Claims vs checks

Vuln counts and Patch Tuesday linkage are Microsoft Security Blog primary. CyberGym-style “topped” is a vendor-cited bench — treat as Microsoft’s snapshot, not an independent SOC audit of residual Windows risk. Later June Build follow-ups cited further score gains (secondary to this May story).

Limits

  • Limited private preview — not broadly available at announcement.
  • Ensemble performance depends on model mix and harness; not transferable as a single-model Elo.
  • Competitive bar vs OpenAI Daybreak / Anthropic Glasswing is desk framing.

Sources

  • Microsoft Security Blog: “Defense at AI speed: Microsoft’s new multi-model agentic security system tops leading industry benchmark” (May 12, 2026).
  • Microsoft Security follow-ups on MDASH preview/internal use (June 2026).
  • Independent technical recaps of the May 12 CVE cohort and MDASH pipeline (May–June 2026).

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading