What Canadian Artificial Intelligence Safety Institute does
The Canadian Artificial Intelligence Safety Institute (CAISI) is a Government of Canada initiative that supports research and practical tools focused on AI safety risks from advanced AI systems—especially risks related to synthetic content (including impersonation and fraud) and risks from systems that may be dangerous or hinder human oversight. CAISI’s goal is to advance the science of AI safety in collaboration with international partners so that governments are well positioned to understand and act on risks as they arise.
CAISI’s work is organized around (1) applied and investigator-led research funded through a research program at the Canadian Institute for Advanced Research (CIFAR) and (2) government-directed projects that support government priorities on AI safety, including joint testing with international counterparts and guidance for AI developers, users, and the public.
Institutionally, CAISI is led by Innovation, Science and Economic Development Canada (ISED) and leverages research capabilities of the National Research Council of Canada (NRC) and Canada’s broader AI research ecosystem through CIFAR and three national AI institutes (Mila, Amii, and Vector Institute).
CAISI also positions itself within Canada’s international AI safety role by coordinating with partners through an International Network of AI Safety Institutes, whose establishment and convenings are referenced as part of CAISI’s global role.
In addition to funding and convening research, CAISI publishes public guidance content aimed at improving how AI evaluations are reported and understood (for example, best practices for AI evaluators).