BRK319 Breakout Tue 17 Nov 16:30 PST
Find and fix code vulnerabilities faster with Codename MDASH
John Nix, Henrique Oliveira
Summary
From the recording / Microsoft AI summary
MDASH is a multi-model agentic code scanner in which more than 100 specialised agents discover, debate, validate and deduplicate vulnerabilities. It is in private preview for eligible organisations and already used inside Microsoft. Speakers said it found 1,415 CVEs in October 2026, more than the 1,273 fixed in all of 2025, and surfaced a FreeBSD remote code execution bug undiscovered for over 20 years. A Defender demo connected GitHub repositories, produced validated findings and opened remediation pull requests via GitHub Copilot. It complements SAST, DAST and secret scanning. A deep-scan mode is coming to GitHub Advanced Security. Customer code is scanned on an encrypted temporary copy and not used to train public models.
Session description
Microsoft's catalogue description
Accelerate vulnerability discovery with multi-model agentic security. Codename MDASH orchestrates specialized AI agents and security-focused AI models to identify, validate, and prioritize vulnerabilities, helping teams understand risk and focus on the issues that matter most.
Demo timestamps
Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.
- Skip to 10:09
Caldova pharmacy app and AI agent workflows
Live demo of the Caldova sample pharmacy app and AI agent workflows. The speaker moves into a dashboard by 11:24, then uses the Defender portal to review findings and start a scan.
- Skip to 12:10
Chained vulnerabilities and broken authorization exploit
Attacker exploit demonstration using chained vulnerabilities and broken authorization against the sample application.
- Skip to 18:38
Creating a remediation pull request from the Defender portal
After MDASH produces validated findings, the speaker creates a pull request from the Defender portal. MDASH can also bring findings back into developer workflows.