The Secure and Private AI (SPY) Lab conducts research on the security, privacy and trustworthiness of machine learning systems. We often approach these problems from an adversarial perspective, by designing attacks that probe the worst-case performance of a system to ultimately understand and improve its safety.

Visit our GitHub organization Twitter account

News

Sep 11, 2024

Our lab member Javier Rando is co-organizing the LLMail Inject competition at SaTML 2025 on adaptive attacks against prompt injection defenses.

Jul 27, 2024

Our papers Stealing part of a production language model and Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining obtained best paper awards at ICML 2024.

Apr 5, 2024

The paper Evading Black-box Classifiers Without Breaking Eggs has been awarded as Distinguished Paper Runner-Up at IEEE SaTML 2024.

Mar 12, 2024

We have reverse-engineered the (secret) Claude 3 tokenizer by inspecting the generation stream. Check our blog post, code and Twitter thread.


People


Avatar

Daniel Paleka

PhD Student

Avatar

Javier Rando

PhD Student

Avatar

Michael Aerni

PhD Student

Avatar

Jie Zhang

PhD Student

Avatar

Kristina Nikolic

PhD Student