About me
I am presently a computer science graduate student at Cornell University. My research interests lie broadly in building trustworthy computer systems. Problems relating to formal specifications, security, extensibility and programmability appeal to me in particular. I am fortunate to be advised by Andrew Myers and Robbert van Renesse to pursue my interests. Before making the questionable decision to become a grad student, I had a blast as an undergrad at IIT Roorkee, where I was awarded the Director's Gold Medal. On the internet, my accounts usually have a character from the greatest show ever.
Research
We can only build secure and trustworthy systems if we know what secure and trustworthy means. In a nutshell, my research is about finding and enforcing the right specifications for computer systems. I'm currently working on the following projects:
Real–ideal semantic foundations for language-based security
Existing security properties such as noninterference are too simplistic; they cannot capture the security requirements of practical systems, which often have intertwined confidentiality and integrity requirements. Our insight is to adapt the real–ideal paradigm from cryptography to give simple and expressive security specifications for computer systems. This has been quite successful. Using this paradigm, we have been able to provide simple semantic definitions of important—but less understood—security properties (such as nonmalleable information flow, robust declassification, and confused deputy attack security). In fact, we show that even syntactic type soundness theorems for information-flow type systems are presented best as real–ideal statements.
Joint work with Andrew Myers, Silei Ren, and Hanxi Chen.
Enforcing auditable specifications for AI prompts
Using AI is exhilarating. Its lack of guarantees is also terrifying. The right interaction model with AI must be to combine LLM prompting with techniques from programming languages and formal methods. We are designing simple, easily auditable, refinable, and enforceable specification languages to constrain AI execution to give users functionality and security guarantees. Moreover, AI can itself generate such specifications, lifting significant user burden; a user only needs to audit the generated specification. Additionally, AI's remarkable ability to write formal proofs can even automate specification refinement! Such an interaction model is an important step in a future where we use our AI superpowers more responsibly. Because, with great power comes great responsibility.
Joint work with Andrew Myers, Alexandra Silva, and Guy Amir.
Self-managing chain replication
The original chain replication protocol has been in use in many companies and research projects. One downside of chain replication is that it cannot stand on its own and requires a configuration service. It turns out that this is not fundamental: a chain replicated service can reconfigure itself, for example after a failure. The new protocol also addresses certain inefficiencies in the original protocol, such as latency increasing with chain length.
Joint work with Robbert van Renesse, Fred Schneider, and Yacqub Mohamed.
Papers
- [In submission] Demystifying Information Flow Downgrading using the Real–Ideal Paradigm - Suraaj Kanniwadi, Andrew Myers
- [In submission] Type Confusion Security - Silei Ren, Hanxi Chen, Suraaj Kanniwadi, Andrew Myers
- [In submission] CR²: Chain Replication Revisited - Suraaj Kanniwadi, Yacqub Mohamed, Fred B. Schneider, Robbert van Renesse
- [USENIX Security '23] Side-Channel Attacks on Optane Persistent Memory - Sihang Liu, Suraaj Kanniwadi, Martin Schwarzl, Andreas Kogler, Daniel Gruss, Samira Khan
Teaching
I've been involved with teaching the following courses at Cornell:
- Fall 2026: Head teaching assistant, CS2800 (Mathematical Foundations of Computing)
- Fall 2025: Head teaching assistant, CS2800 (Mathematical Foundations of Computing)
- Fall 2024: Guest lecture, CS4450 (Introduction to Computer Networks)
- Spring 2024: Guest lecture, CS4450 (Introduction to Computer Networks)
- Fall 2023: Head teaching assistant, CS3410 (Computer System Organization and Programming)
- Spring 2023: Head teaching assistant, CS4450 (Introduction to Computer Networks)
- Fall 2022: Head teaching assistant, CS4450 (Introduction to Computer Networks)
Talks
- 27 Oct 2026: Why security? What security? How security? [Cornell Tech. AI4AI Fall Retreat] [slides]
- 26 Aug 2026: Demystifying Information Flow Downgrading using the Real–Ideal Paradigm [PLDG @ Cornell] [slides]
- 16 Apr 2025: The Real–Ideal Paradigm for Systems [Cornell A-Exam] [slides]
- 07 Feb 2025: An Algebraic Theory of Systems [Cornell Systems Lunch] [slides]
- 13 Sep 2024: Verifying Hardware Security Modules with Information-Preserving Refinement [Systems Lunch @ Cornell] [slides]
- 28 Feb 2024: Secure Synthesis of Distributed Cryptographic Applications [Cornell PLDG] [slides]
- 11 Aug 2023: Side-Channel Attacks on Optane Persistent Memory [USENIX Security '23] [slides]
- 16 Feb 2023: Hyperproperties [Cornell CS 6156] [slides]
- 27 Sep 2022: Paxos made Simple [Cornell CS 6410] [slides]
Software
My GitHub account has a bunch of software I've written over the years. This includes my vim configuration and many plugins. There's also some software that I use to manage my swimming. The source for this website is here.