This paper explains some of the challenges involved in protecting the system memory of a computing platform against adversaries who can read and modify its contents. To clarify the limitations and effectiveness of different defence mechanisms, we lay out a threat model, coupled with a set of assumptions on the adversarial capabilities, for a variety of scenarios. Finally, we explain the mechanics of some attacks on encrypted memory, to illustrate why relying on encryption alone is risky in reality. The paper is addressed to individuals and organisations with a need to consider physical attacks as part of their threat scope and to companies considering solutions to isolate system administrators from customer data in virtualised environments.