Canada and Germany Invest in LawZero to Develop Safe, Sovereign AI
At the ALL IN conference in Montréal, Canada and Germany announced joint investments totaling CAD 150 million and EUR 100 million respectively into LawZero. The non-profit, founded by Yoshua Bengio, aims to create 'Scientist AI,' a transparent system designed without autonomous goal-seeking behaviors.
On September 16, 2026, during the ALL IN conference in Montréal, the Governments of Canada and Germany announced a significant joint investment in LawZero, a non-profit organization founded by Turing Award winner and Université de Montréal professor Yoshua Bengio. The funding is intended to support the development of "Scientist AI," an artificial intelligence system designed with safety as a core principle.
The Government of Canada will contribute CAD 150 million through its Strategic Response Fund. Simultaneously, the Federal Government of Germany plans to invest EUR 100 million, a commitment that remains subject to notification by the European Commission. This initiative seeks to strengthen technological sovereignty for both nations by reducing reliance on foreign-controlled AI systems and establishing dedicated sovereign computing infrastructure in partnership with Hypertec and 5C.
LawZero, which was established approximately one year prior to this announcement and employs close to 50 people at its Montréal headquarters, is developing a model distinct from current mainstream approaches. The "Scientist AI" project focuses on creating a system that reasons transparently and avoids autonomous goal-seeking behaviors associated with reinforcement learning. Yoshua Bengio cited concerns about AI safety, including recent incidents involving OpenAI agents hacking Hugging Face, as motivation for pursuing this new approach.
Beyond the technical objectives, the Canadian portion of the investment is projected to create 360 full-time jobs within Canada. The collaboration also aims to deepen bilateral cooperation between Canada and Germany on AI regulation and development, advancing models that avoid harmful traits such as deception and sycophancy.