Abstract

We are interested in addressing a series of existing and plausible threats to cybersecurity where the adversary possesses unconventional attack capabilities. Such unconventionality includes, in our exploration but not limited to, crowd-sourcing, physical/juridical coercion, substantial (but bounded) computational resources, malicious insiders, etc. Our studies show that unconventional adversaries can be counteracted with a special anchor of trust and/or a paradigm shift on a case-specific basis.

Complementing cryptography, hardware security primitives are the last defense in the face of co-located (physical) and privileged (software) adversaries, hence serving as the special trust anchor. Examples of hardware primitives are architecture-shipped features (e.g., with CPU or chipsets), security chips or tokens, and certain features on peripheral/storage devices. We also propose changes of paradigm in conjunction with hardware primitives, such as containing attacks instead of counteracting, pretended compliance, and immunization instead of detection/prevention.

In this thesis, we demonstrate how our philosophy is applied to cope with several exemplary scenarios of unconventional threats, and elaborate on the prototype systems we have implemented. Specifically, Gracewipe is designed for stealthy and verifiable secure deletion of on-disk user secrets under coercion; Hypnoguard protects in-RAM data when a computer is in sleep (ACPI S3) in case of various memory/guessing attacks; Uvauth mitigates large-scale human-assisted guessing attacks by receiving all login attempts in an indistinguishable manner, i.e., correct credentials in a legitimate session and incorrect ones in a plausible fake session; Inuksuk is proposed to protect user files against ransomware or other authorized tampering. It augments the hardware access control on self-encrypting drives with trusted execution to achieve data immunization. We have also extended the Gracewipe scenario to a network-based enterprise environment, aiming to address slightly different threats, e.g., malicious insiders.

We believe the high-level methodology of these research topics can contribute to advancing the security research under strong adversarial assumptions, and the promotion of software-hardware orchestration in protecting execution integrity therein.