Canada and Germany fund Bengio's safe AI push

Canada and Germany plan to put CAD 150 million and EUR 100 million into LawZero. The Montreal non-profit is building a safe-by-design alternative to autonomous AI agents, according to the announcement made September 16 at the ALL IN conference in Montreal.

It is a sovereignty bet as much as a safety bet.

LawZero was founded and is scientifically led by Turing laureate Yoshua Bengio. It is building Scientist AI, which differs from frontier systems trained to act autonomously and pursue objectives. Instead, Scientist AI is designed to reason transparently and produce evidence-based outputs with no goals of its own. Canada says its share, routed through the Strategic Response Fund, will pay for research and engineering talent and computing capacity at home.

The timing is what turns this from a simple grant into something bigger. On June 4, 2026, Canada launched its National Artificial Intelligence Strategy, AI for All. It made safety and security central and promised stronger national AI safety capabilities. In February, Canada and Germany signed a Joint Declaration of Intent on AI and launched the Sovereign Technology Alliance. September 16 is the first capital call on that framework, with ministers Evan Solomon and Germany's Karsten Wildberger announcing the plan.

Ottawa ties the money to hardware and jobs. The project will create 360 full-time jobs in Canada and build dedicated sovereign computing infrastructure with Canadian firms Hypertec and 5C. That anchors talent and compute domestically and reduces reliance on foreign-controlled systems. Berlin's EUR 100 million will fund R and D at LawZero's new German office and is meant to incorporate European norms from the outset.

There is a catch. Germany's funding is still subject to notification by the European Commission, so it is not yet final. And LawZero is not promising a frontier replacement tomorrow. The near-term work is tools to help assess and oversee existing AI systems and to support scientific research, before advancing toward safe-by-design frontier models.

The guardrail ambition is explicit. LawZero positions Scientist AI as an objective scientist that works purely on facts, not an agent with preferences. That includes acting as a check on AI agents that show deceptive or self-preserving behaviour, such as trying to avoid being turned off.

If the build lands, Canada and Germany get domestically anchored labs, compute and evaluation tools for AI that can explain itself. If it slips, they will have funded a long-horizon safety methodology while frontier deployment keeps accelerating elsewhere.