The Canadian and German governments are committing up to $300 million in grant funding to LawZero, the non-profit founded by Turing Award winner Yoshua Bengio to build AI systems designed to avoid the failure modes of current models1,2. Each country will contribute up to $150 million.
The funding represents a major sovereign bet on safety-first AI development at a moment when, as Bengio put it, "Safety has become very important in the eyes of the public, for good reason. We urgently need to have a plan to build AI that is not going to do really bad, misaligned things".
What LawZero is building
Bengio, a professor at the Université de Montréal and former scientific director at Mila, founded LawZero last year with close to US$30 million in philanthropic funding after growing concerned that powerful AI systems could be misused or slip from human control. The organization employs close to 50 people and is based in Montreal.
LawZero is developing a system it calls Scientist AI, which the organization says will not exhibit the negative traits found in today's models. The first priority for Scientist AI is to build a guardrail that would act as a monitor of other AI systems and prevent agents from taking actions that could cause harm.
The new financing from Canada and Germany will fund additional hiring and the computational processing costs associated with this research.
Broader safety momentum
The grant arrives as calls for coordinated restraint in AI development gain volume. Bengio was one of the signatories to an open letter in 2023 arguing for a pause on AI development. Anthropic CEO Dario Amodei wrote an essay this month arguing that AI companies should work together to pace the speed of development.
LawZero is in talks with other countries about potential additional funding.
ANALYSIS The $300 million commitment channels public money not into a national lab or a commercial champion but into a non-profit explicitly chartered around safety constraints, a structure that separates LawZero from the compute-scaling race among frontier labs. The focus on building a monitoring guardrail for other AI systems, rather than a competing general-purpose model, positions the effort as infrastructure for the broader ecosystem rather than a direct commercial rival.