Canada and Germany fund Yoshua Bengio’s lab as tech sovereignty bets on an artificial intelligence with no specific goals

In Montreal, two governments decided to allocate public money to build an artificial-intelligence model that wants nothing at all. On 16 September 2026, Canada and Germany announced joint funding for a nonprofit founded less than a year ago, hoping to show that the greatest risk in modern models is not their weakness but their desire to please you.

The two amounts were announced officially: 150 million Canadian dollars from Ottawa and about 100 million euros from Berlin. The beneficiary is LawZero, the organization founded by deep-learning researcher Yoshua Bengio, Turing Award winner and professor at the University of Montreal, who has become since 2023 one of the most urgent voices warning about the risks of advanced systems.

A model that understands the world without pursuing it

The system being built is called Scientist AI, and its central idea is simple to describe but difficult to implement: to represent the world as neutrally as possible, to predict and evaluate hypotheses without pursuing its own specific goals. The difference here is not cosmetic. Most of what we use today is trained to please the user and accomplish a task, which is precisely what makes its behavior hard to predict when the task becomes complex. This model, by contrast, is intended to be a witness rather than an actor, and its first expected uses are to assess existing systems, not to replace them.

Why two governments pay the price of caution

Here the real lens of the story appears, which is sovereignty. Canadian AI minister Ivan Solomon said that investing in local innovation such as this system strengthens his country’s sovereign capacity in artificial intelligence and ensures that the next generation is built safely. German digitisation minister Carsten Feldberg linked the matter explicitly to “technological sovereignty” by building advanced research capabilities jointly in Canada and Europe. The official Canadian statement goes further: anchoring research, talent and computing infrastructure within the country reduces reliance on systems controlled from abroad.

The implementation details reveal that the talk is not merely rhetorical. The project is expected to create 360 full-time jobs in Canada and to build sovereign computing infrastructure with two local companies, Hypertec and 5C. In other words, the two countries are not buying a research outcome but a position within the production chain itself.

The announcement builds on a bilateral cooperation between the two countries that started last February, making it more than a standalone initiative. When two governments outside the US-China bloc come together on a single project, the announcement is not merely funding but an attempt to secure a third seat in a discussion that usually proceeds without them.

What this means for those building sovereignty in the region

For a reader in Riyadh, Abu Dhabi or Cairo, the message is sharper than it appears. Talk of digital sovereignty in our region is often translated into data centres, chips and computing capacity, which are undoubtedly necessary conditions. But what Ottawa and Berlin have done is purchase a share in the hardest layer: defining how the model should behave in the first place. Whoever controls the ability to evaluate models practically controls the standards for their acceptance, and anyone who merely imports the model and the standard together remains a consumer no matter how many megawatts it consumes.

What remains unresolved

It must be said plainly: the project is still in the research phase. The papers published so far address partial concepts and do not provide comprehensive safety guarantees, and the German funding itself requires notification to the European Commission before its path is completed. In other words, what has been announced is a funded bet, not a ready product, and it may fail.

Even if the implementation stumbles, the political signal has been sent: two governments have decided that model safety is not an ethical file left to major labs, but a sovereign spending line incorporated into the budget. That alone changes the question to ask. It is no longer when we will get the model, but who determines when the model is fit for use for us.