AI Safety Logo

Reliability and Robustness of Foundational Models

The reliability and robustness of AI-powered apps do not only depend on the traditional software security methods, but also on the security of the underlying AI models. In this area, we explore the vulnerabilities of foundational models such as jailbreaks, hallucinations, and unsafe code generation among others, and devise new defense mechanisms. With our methods, we hope to make AI-powered applications safer to use.

Publications