Amii: AI trust & safety
Standing Trust & Safety programme. Amii's own programme page states "Amii's Trust & Safety program is advancing the science of AI safety" and, as of the January 2026 year-in-review, "We established a five-person team including program leadership, two Applied Research Scientists, a Machine Learning Engineer, and a postdoctoral fellow" with an agenda "to study vulnerabilities in frontier models and explore how new safeguards can be designed, tested, and evaluated". The research builds on Amii's reinforcement-learning strength; the one paper on its Trust and Safety Research index, "Learning to be Cautious", studies RL agents that learn cautious, robust policies under reward uncertainty. CAISI partner alongside CIFAR, the NRC, Mila and Vector.