In Windows 10, Microsoft is introducing a radical new concept to the underlying OS architecture, and likely the biggest change to the NT design since the decision to move the GUI in kernel-mode.
In this new model, the Viridian Hypervisor Kernel now becomes a core part of the operating system and implements Virtual Secure Machines (VSMs) by loading a true microkernel - a compact (200kb) NT look-alike with its own drivers called the Secure Kernel Mode (SKM) environment, which then uses the Hypervisor to hook and intercept execution of the true NT kernel. This creates a new paradigm where the NT Kernel, executing in Ring 0, now runs below the Secure Kernel, at Ring ~0 (called Virtual Trust Level 1).
But it doesn't stop there - as the Ring 0 NT kernel now has the ability to not only create standard Ring 3 user-mode applications, but also Ring ~3 applications (or Virtual Trust Level 0) that run in Isolated User Mode (IUM). Because VTLs are all more privileged than Ring 0, this now creates a model where a user-mode application running inside a VSM now has data and rights that even the kernel itself cannot modify. Why go through all this trouble? Because it seems like the hottest thing these days is Pass-the-Hash, and attacks must seemingly be mitigated at all costs. And even in Windows 8.1, an attacker with the permissions to load a kernel driver can bypass the existing mitigations (and Mimikatz is signed!). With VTLs, now even the most privileged attacker is only as privileged as the hypervisor will allow it - never able to truly read the hash date that is stored in the secure partition.
How "secure" is this new model really? And what prevents a malicious application from running in such a secure mode to begin with?