On the first page of the Introduction, it's clear that the ME should not have access to the actual key in TME. In MKTME, there seems to be one key that is held

On the first page of the Introduction, it's clear that the ME should not have access to the actual key in TME. In MKTME, there seems to be one key that is held only by the CPU while the others can be provided by software. I think the use case for this is a) to defend runtime application memory on systems that use non-volatile memory as RAM (e.g., 3D XPoint), b) to defend against cold boot attacks, and c) to defend against attacks where one VM can read the data pages of another VM, possibly through an re-assignment of a previously used and uninitialized memory page.