Colin Toft

About Me
Hi, I'm Colin. I've always loved learning how things work — growing up that meant juggling, Rubik's cubes, and card tricks. These days it means exploring everything from low-level systems and compilers to AI safety and alignment.
I'm currently an Anthropic Fellow focused on red teaming for AI control, where I study whether current monitoring systems can reliably detect unintended model behavior. I'm also finishing up my Computer Science degree at the University of Waterloo.
Previously, I've worked on GPU compilers at NVIDIA, ML for autonomous driving at Waabi, and co-authored a paper on AI-enabled compiler optimization (ACM/IEEE CASES 2024).
Outside of research, I build software for nonprofits at UW Blueprint, work on creative projects at Socratica, write and produce music, and stay active through running and rock climbing.
Projects
UW Blueprint - Sistema Toronto
Created a platform for registered charity Sistema Toronto, which provides music lessons for students in underserved communities.
Idle Motion in Human-Robot Interaction
Undergraduate research assistantship at the Social and Intelligent Robotics Research Lab (SIRRL).
UW Blueprint - Feeding Canadian Kids
Created a full-stack platform for Feeding Canadian Kids charity, connecting food providers with after-school programs.