Microsoft has introduced a multi-model AI system called MDASH, designed to enhance vulnerability discovery and remediation processes. Currently in limited private preview testing with select customers, MDASH employs over 100 specialized AI agents for various classes of vulnerabilities, enabling autonomous discovery, validation, and demonstration of exploitable defects in complex codebases. The system operates through a structured pipeline that analyzes source code, constructs threat models, and validates findings using auditor and debater agents. MDASH has successfully identified 16 vulnerabilities in its initial tests, including two critical flaws affecting Windows networking and authentication:
1. CVE-2026-33824 (CVSS score: 9.8) - A double-free vulnerability in "ikeext.dll" allowing remote code execution via specially crafted packets.
2. CVE-2026-33827 (CVSS score: 8.1) - A race condition vulnerability in Windows TCP/IP ("tcpip.sys") enabling remote code execution through specially crafted IPv6 packets.