Ten million Canadian dollars for the laboratories that birthed deep learning

Anthropic funds eight Canadian partners with ten million dollars. What this support for research laboratories changes for AI safety.

Anthropic is announcing a ten-million Canadian dollar commitment to research institutions in the country, coupled with eight partnerships. The funding is intended to support work on beneficial and responsible applications of AI.

The choice of Canada is built on a historical lineage that the company highlights: at a time when neural networks faced widespread skepticism, the University of Toronto and the Université de Montréal were among the few institutions to support this path, while the University of Alberta laid the groundwork for reinforcement learning.

The partners span three categories. First, the regional institutes—Amii in Edmonton, Mila in Montreal, and the Vector Institute in Toronto—which will work on reinforcement learning, trust and safety, healthcare, and science. Second, healthcare, with CHEO, focused on pediatrics, and the Centre for Addiction and Mental Health (CAMH), whose Krembil Centre for Neuroinformatics aims to evaluate the fairness of AI systems applied to psychiatry. Finally, universities: Université Laval will study the behavior of language models across cultural contexts and low-resource languages, including Quebec French and Indigenous languages; the University of Saskatchewan will cover biomedicine, food security, and quantum computing; and the University of Toronto will fund projects through a scientific review process. Additionally, Amii, Mila, and Vector will join the company's startup program this summer.

The announcement is accompanied by a country profile drawn from the company's in-house economic index, which ranks Canada eighth globally in Claude usage, with per capita adoption higher than its population size would suggest. One detail stands out: translation requests peak in provinces where the public service is most prominent, a likely effect of federal bilingualism requirements.