Canada and Germany will put up to CAD $300 million into Yoshua Bengio's safety non-profit
The plan is a model that reasons in the open and chases nothing, and its first job is watching the frontier models that already exist.
On September 16, 2026, at the ALL IN conference in Montréal, Canada and Germany said they plan to pay for an AI built to hold no goals of its own. Canada is putting in CAD 150 million. Germany is putting in EUR 100 million.
The money goes to LawZero, the not-for-profit Yoshua Bengio founded, which puts the joint commitment at up to CAD $300 million. Granted, some of that is still a plan rather than a payment.
- Announced
- September 16, 2026, at ALL IN
- From Canada
- CAD 150 million, through the Strategic Response Fund
- From Germany
- EUR 100 million, subject to notification by the European Commission
- Recipient
- LawZero, headquartered in Montréal
- Jobs
- 360 full-time jobs in Canada
What the money buys
Canada's half runs through the Strategic Response Fund and pays for research and engineering talent and for computing capacity. The Canadian release counts "360 full-time jobs in Canada" and a sovereign computing setup built with Hypertec and 5C. Germany's half pays for a new LawZero office in Berlin.
Read down that release and it turns out the German half carries a condition (it is "subject to notification by the European Commission"). The two ministers who announced the money were Canada's Evan Solomon and Germany's Karsten Wildberger.
What Scientist AI is supposed to be
What surprised me is how flatly the design brief contradicts the rest of the field. LawZero sets its work against what it calls current frontier AI systems, the ones built to act on their own and chase their own goals. I think that contrast is the news here, more than the number is.
Unlike current frontier AI systems designed to act autonomously and pursue their own goals, Scientist AI is being developed to reason transparently and provide reliable, evidence-based outputs unbiased by goals of its own.
Scientist AI is meant to work on the models that already exist before it is anything else. LawZero says it expects the approach to provide guardrails for today's frontier models and to speed up scientific work (the release treats reliability as a core capability, which is a different starting point from making a model capable and then fixing it).
Bengio's own line in the release is short.
As AI risks multiply and accelerate, our priority must be building solutions to make this technology safe
Why two governments paid for it
Ottawa frames the spending as sovereignty. Anchoring the research and the computing in Canada will "strengthen Canada's technological sovereignty and reduce reliance on foreign-controlled AI systems", the release says, and the project builds on the Joint Declaration of Intent on Artificial Intelligence and the Sovereign Technology Alliance the two countries announced in February.
Germany's interest is written into the money. Its funding covers the work at the Berlin office and, the Canadian release says, makes sure European norms and regulations in the field of AI "can be incorporated from the outset of the project".
LawZero's work begins with tools to help assess and oversee existing AI systems and to support scientific research. Germany's EUR 100 million is subject to notification by the European Commission.