Sai Praneeth Karimireddy

Sai Praneeth Karimireddy

Assistant Professor of Computer Science (& by courtesy, ECE), USC

I work on principled approaches to trustworthy AI — combining optimization, statistics, and economics to design, evaluate, and improve real-world AI systems. Before USC I was an SNSF postdoc with Michael I. Jordan at UC Berkeley, and I completed my PhD at EPFL advised by Martin Jaggi. My work has been deployed at Meta, Google, OpenAI, and Owkin. See the group page for how to work with me.

Research

AI is changing the world in unprecedented ways, turning once-philosophical questions — what is truth, what improves human welfare — into urgent technical ones. I take a principled approach to formalizing, understanding, and answering them. Some things I'm thinking about:

Trustworthy AI. What does it mean for LLMs to be safe or private? How can we evaluate and characterize LLM capabilities and behaviors?

AI ecosystems. How will AI agents interact with humans, society, and each other? How can we understand and shape the resulting emergent phenomena?

AI in health & finance. Can we translate these insights to deploy AI in high-stakes settings such as healthcare and finance?

Selected Work

Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment
ICML 2026

What are the fundamental limits of black-box auditing?

EPSVEC: Efficient and Private Synthetic Data Generation via Dataset Vectors
ICML 2026

How do you feed context to an LLM in a differentially-private manner?

Mechanisms that Incentivize Data Sharing in Federated Learning
Arxiv 2022🏆 Best paper

How do you model multi-agent cooperation and design incentives without monetary rewards?

Byzantine-Robust Learning on Heterogeneous Datasets via Bucketing
ICLR 2022🏆 Spotlight

What distinguishes important corner-case data from noisy outliers?

News

Jun 2026

Serving as an Area Chair at NeurIPS 2026.

Mar 2026

Gave a talk at the Google Privacy in ML seminar on privacy-preserving synthetic data generation (video to come).

Feb 2026

Gave a talk to Capital One on Private Synthetic Datasets for Enterprise AI. Slides.

Jan 2026

Appointed as visiting researcher at Simons Institute, UC Berkeley. Reach out if you are in the Bay Area and want to chat!

Jan 2026

Gave a tutorial on Data Valuation and incentives for data sharing, together with Han Shao.

Nov 2025

Invited to give a talk at The Nexus of Open Science event in Washington, DC.

Show more news
Sep 2025

Serving as an Area Chair at ICLR 2026.

Jun 2025

Excited to help run the workshop on Incentives in Data Sharing at TTIC this summer — submissions and registration now open.

Feb 2025

Co-organizing a TTIC summer workshop on Incentives in Data Sharing and Collaborative Learning (Aug 13–15). Reach out if you would like to give a talk or take part.

Jan 2025

Named a Capital One Fellow along with Robin Jia — thank you for the support!

Dec 2024

Received a joint appointment with the Ming Hsieh Department of ECE. You can now apply to work with me through PhD programs in both CS and ECE.

Nov 2024

Gave talks in the Chicago area — the CS seminar at the University of Chicago and the LANS seminar at Argonne National Laboratory.

Sep 2024

Invited talks at the USC Theory Lunch and INFORMS 2024, plus a guest lecture in USC CSCI 697 on "Building Collaborative Data-Ecosystems".

Sep 2024

Serving as an Area Chair at ICLR 2025.

Aug 2024

Co-organizing the workshop on Federated Learning in the Age of Foundation Models at NeurIPS 2024.

Jun 2024

Appointed co-lead of the Data Quality and Federated Learning working group with Holger Roth as part of MONAI.

Jun 2024

At the Foundations of Responsible Computing conference (FORC 2024) in Boston, Jun 12–14.

Awards