Marika Swanberg

I am an AI Safety Engineer and Researcher on the Chrome team at Google in New York City. I design and productionize safety mechanisms for AI agents (see this blog post featuring my team's work). I was briefly a part of the Privacy Sandbox where I did in-house differential privacy research for private model training at scale.

Prior to joining Google, I completed my PhD in Computer Science at Boston University, where I was advised by Adam Smith and served as a Hariri Institute Graduate Student Fellow. My research focused on differential privacy , and I also worked on semi-related topics like: measuring pretraining data memorization by LLMs, formalizing the Right to Be Forgotten, and modeling secure mesaging in the universal composability framework.

Persistent themes in my work

• Threat modeling: Carefully defining the powers and scope of adversaries in a variety of settings.
• Designing real-world defenses: Designing effective and scalable defenses for real systems.
• Breaking things: Whether it's a lower bound argument or a defense, I like to find weak points.

Some prior dabblings

• Teaching: I was a Visiting Assistant Professor (and faculty chaperone for the ski club) at Reed College for the Spring 2023 semester, where I taught Algorithms and Data Structures.
• Internships: I was lucky to have interned at Tumult Labs with Damien Desfontaines and Sam Haney, as well as twice at Google Research, on Sergei Vassilvitskii's team and Peter Kairouz's team.
• Editing textbooks: During my undergrad I contributed detailed feedback to Boaz Barak's forthcoming book on Introduction to TCS as well as Irena Swanson's Introduction to Analysis textbook.