Reshabh K Sharma

I am a PhD student at Paul G. Allen School of Computer Science & Engineering . I am advised by Professor Dan Grossman. I work at the intersection of Programming Languages (PL) and Security. My research focuses on the development of infrastructure for developing software based mitigation for a broad range of safety and security vulnerabilities ranging from microarchitectural side channel attacks to high-level systems including Mixed Reality (XR) and Large Language Models (LLMs).
Need for software based mitigations: They provide a flexible and timely response to emerging threats, allowing for updates and patches to be deployed rapidly as vulnerabilities are discovered. They also enhance the overall security posture by layering defenses, protecting systems even when hardware-based solutions fall short or are not feasible.
Need for mitigation development infrastructure: The necessity for a specialized mitigation development infrastructure arises from existing constraints, where developers find it challenging to adequately protect their systems. Mitigations, typically created by vendors for operating systems and compilers, offer limited flexibility, functioning as simple on/off switches without considering the unique challenges of individual applications. Developers, who have a deeper understanding of their applications' specific security needs, lack the tools to custom-build software-based mitigations. There's a crucial need for an infrastructure that enables developers to devise and apply custom defenses, ensuring a higher level of security and safety tailored to each application's demands.

Securing LLM-based systems

Reshabh K Sharma, Vinayak Gupta, and Dan Grossman. Defending Language Models Against Image-Based Prompt Attacks via User-Provided Specifications SAGAI at IEEE S&P. To appear (2024).

Reshabh K Sharma, Vinayak Gupta, and Dan Grossman. SPML: A DSL for Defending Language Models Against Prompt Attacks. arXiv preprint arXiv:2402.11755 (2024)

Securing Low-Level systems

Michael Flanders, Reshabh K Sharma, Alexandra E. Michael, Dan Grossman, and David Kohlbrenner. Avoiding Instruction-Centric Microarchitectural Timing Channels Via Binary-Code Transformations. ASPLOS. To appear (2024).