Two governments just made one of the largest public bets on AI safety research to date. On September 16, at the ALL IN conference in Montréal, Canada and Germany announced joint funding of up to CAD $300 million for LawZero, the not-for-profit AI lab founded by Turing Award winner Yoshua Bengio. According to reporting on the announcement, each country is putting in up to $150 million.

That number lands in a field where safety research has mostly survived on philanthropy. Until now, LawZero had been running on roughly US$35 million in donations from foundations and individual donors. A jump of this size changes what the organization can attempt, and it comes from two national treasuries rather than from a tech company or a donor network.

Where the Money Goes

The funding covers three things the press release spells out directly: growing an international scientific research team, securing computing infrastructure, and opening a European office.

The compute side has a specific shape. LawZero will build dedicated sovereign compute infrastructure inside Canada, working with two Canadian partners, Hypertec and 5C. Hypertec CEO Simon Ahdoot described the arrangement as a partnership between government, capital and Canadian industry, with his company building the infrastructure and 5C providing dedicated capacity. Keeping the hardware on Canadian soil is part of the point. A safety lab that rents capacity from a foreign hyperscaler ends up dependent on the same companies whose models it is trying to supervise.

The European side is a new office in Berlin. That gives LawZero a physical presence inside the EU and connects the Canadian research base to Germany’s funding commitment in a way that goes past a wire transfer.

What Scientist AI Tries To Do Differently

LawZero’s flagship research program is called Scientist AI, and the approach behind it is the reason two governments are willing to write this kind of cheque.

Most frontier AI systems are built to act. They pursue objectives, take actions in the world, and increasingly operate as agents that chain multiple steps together without a human reviewing each one. Scientist AI is being designed around a different premise. LawZero describes it as a system built to reason transparently and produce reliable, evidence-based outputs, without pursuing independent goals. The comparison the organization reaches for is an objective scientist: a system that explains and evaluates rather than one that decides and executes.

German Federal Minister for Digitalization and Government Modernization Karsten Wildberger described the approach as mathematically grounded and framed it as an alternative to autonomous AI agents that could become difficult to control. He also tied it to technological sovereignty for both countries.

The practical near-term use LawZero points to is guardrails. A system that evaluates claims without a stake in the outcome can sit alongside today’s frontier models and flag when those models are behaving in ways their developers did not intend. LawZero also expects Scientist AI to help accelerate scientific work and to improve understanding of where AI risk comes from.

The Timing

The announcement arrives during a stretch where safety questions have moved from conference panels into incident reports. LawZero’s release points to recent cases of AI systems crossing safety boundaries, and coverage of the funding connected it to a broader debate over how much independence advanced systems should be given.

Bengio framed reliability as a capability rather than a constraint. In his statement, he argued that recent events show safety is itself a core capability, and that scientific innovation and public safety can advance together. He called the backing from both governments a defining moment for the organization.

That framing does real work in the funding argument. Safety research has often been sold to governments as a brake on progress, which makes it a hard sell when the same governments are trying to attract AI investment. Presenting a safe-by-design system as a research product with commercial and scientific value shifts the pitch.

How LawZero Got Here

LawZero launched publicly in June 2025 with around US$30 million in philanthropic funding. Bengio told Axios at the time that the amount was expected to support roughly 18 months of basic research. Early donors included Coefficient Giving, the Future of Life Institute and Jaan Tallinn.

In August 2025, the Gates Foundation added a grant supporting work on AI safety benchmarks, algorithms and applications to scientific research and drug discovery, bringing total philanthropic funding past US$35 million. Nvidia and Schmidt Sciences also appear on LawZero’s list of backers.

Canadian government involvement did not start this week. In February 2026, The Globe and Mail reported that Ottawa had signed a letter of intent with LawZero, with a figure above $100 million reportedly under discussion. That same month, Canada and Germany signed a joint AI declaration and launched a Sovereign Technology Alliance, and the ministers involved specifically discussed collaborating with organizations working on safe-by-design systems, naming LawZero.

The organization is governed by an independent board and a Global Advisory Council with an unusual roster for an AI lab: former prime ministers Jacinda Ardern and Stefan Löfven, author Yuval Noah Harari, former Rolls-Royce CEO Sir John Rose, Nike Foundation founder Maria Eitel as board chair, and Mila CEO Valerie Pisano. LawZero was initially incubated at Mila, the Montréal research institute Bengio co-founded.

What Remains Unclear

The commitment is described as up to CAD $300 million, language that usually implies milestones rather than a lump sum. Neither government has published a disbursement schedule, and the release does not say over how many years the money flows.

There is also no public timeline for Scientist AI itself. LawZero has described the research direction and the intended function, but has not announced a system, a benchmark result, or a date. For a lab that has now been handed national-scale funding by two countries, that gap is the thing to watch over the next year.