Microsoft's MDASH AI Finds 16 New Windows Flaws

Microsoft has launched an AI-powered vulnerability detection system called MDASH that identified 16 previously unknown flaws in Windows, including four critical remote code execution vulnerabilities patched as part of the company's May Patch Tuesday updates.

The multimodel agentic scanning harness, developed by Microsoft's Autonomous Code Security team and Windows Attack Research and Protection group, orchestrates more than 100 specialized AI agents across frontier and distilled models. These agents handle a multistage pipeline, from scanning source code for potential issues to validating findings, deduplicating results, and generating proof-of-concept inputs for human review. Microsoft described the system as model-agnostic, allowing flexibility in swapping underlying AI technologies.

Among the vulnerabilities, two stood out for their potential impact. CVE-2026-33827 involves a remote unauthenticated use-after-free flaw in the Windows TCP/IP stack, exploitable via specially crafted IPv4 packets using the Strict Source and Record Route option. CVE-2026-33824 is a preauthentication double-free issue in the IKEEXT service, affecting RRAS VPN, DirectAccess, and Always On VPN...