Security, Privacy, Cryptography × Machine Learning

Harish
Karthikeyan

Senior Research Scientist
J.P. Morgan AI Research & AlgoCRYPT Center of Excellence
277 Park Avenue, New York, NY  ·  he / him / his

Designing, proving, and building privacy-enhancing technologies for AI/ML — secure aggregation for federated learning, lattice-based and zero-knowledge protocols, and privacy for AI agents.

Publications Get in Touch
Portrait of Harish Karthikeyan
Ph.D. NYU ’22
M.S. Columbia ’16
B.Tech NIT Trichy ’15

Updates

News & Announcements

2026 New Gold Reviewer Award, ICML 2026, for being among the top 25% of reviewers.
2026 New TAPAS: Efficient Two-Server Asymmetric Private Aggregation Beyond Prio(+) accepted at Theory and Practice of Multi-party Computation (TPMPC) 2026
2026 New Invited to give a talk on Secure Aggregation and Federated Learning at Flower AI Summit, London, 2026.
2026 Serving on the Program Committee of NeurIPS 2026, ICLR 2027, and AAAI 2027.
2026 Co-organizing the Privacy-Preserving Machine Learning (PPML) affiliated event at CRYPTO 2026 (also 2024 and 2025).
2026 Serving on the Program Committees of ACM CCS 2026 and Financial Cryptography 2027.
2026 Updatable Public-Key Encryption from Class Groups: New Constructions with Better Efficiency accepted at Financial Cryptography 2026, Saint Kitts & Nevis.
Dec 2025 AgentCrypt and HashMark presented at NeurIPS 2025 Workshops.
Oct 2025 Armadillo — single-server secure aggregation with robustness against malicious clients — presented at ACM CCS 2025, Taipei.
2025 OPA: One-shot Private Aggregation with Single Client Interaction accepted at CRYPTO 2025.

Speaking

Selected Talks

2026 Secure Aggregation and Federated Learning — invited talk, Flower AI Summit, London.
2026 TAPAS: Efficient Two-Server Asymmetric Private Aggregation Beyond Prio(+) — contributed talk, TPMPC 2026. Slides

Focus Areas

Research Interests

01 — Privacy-Preserving FL

Federated Learning & Secure Aggregation

Scalable, post-quantum secure aggregation — one-shot and two-server protocols, input validation against malicious clients, communication-efficient compression, and differential privacy.

02 — Cryptographic Protocols

Applied & Post-Quantum Cryptography

Lattice-based zero-knowledge proofs, updatable and forward-secure encryption, verifiable noise for private statistics, and security proofs for protocols that get deployed.

03 — AI Security

Secure AI Agents & Data Provenance

Cryptographic privacy and authorization for multi-agent LLM systems (AgentCrypt), and hash-based watermarking to track the provenance of synthetic training data (HashMark).

Biography

About

I am a Senior Research Scientist at J.P. Morgan AI Research and the J.P. Morgan AlgoCRYPT Center of Excellence. My research sits at the intersection of cryptography and machine learning, focused on privacy-enhancing technologies for AI/ML systems.

I earned my Ph.D. in Computer Science in 2022 from New York University under Yevgeniy Dodis, with a dissertation titled Cryptography: From Practice to Theory. During doctoral studies, I collaborated with Antigoni Polychroniadou on privacy-preserving consensus.

Previously, I received an M.S. from Columbia University (with Allison Bishop) and a B.Tech from NIT Trichy, where I received the Institute Gold Medal.

I co-organize the Privacy-Preserving Machine Learning (PPML) event affiliated with CRYPTO, serve on program committees for ACM CCS, Financial Cryptography, NeurIPS, ICLR, and AAAI, and mentor with Out in Tech and Queer in AI.

Education

Doctor of Philosophy, 2022

New York University, Courant Institute

Advisor: Yevgeniy Dodis · Thesis

Master of Science, 2016

Columbia University

Advisor: Allison Bishop

B.Tech, 2015 — Institute Gold Medal

National Institute of Technology Trichy

Advisor: Kunwar Singh