The German government is setting a new course for artificial intelligence safety and sovereign infrastructure. At the ALL IN conference in Montreal, the German Federal Ministry for Digital and State Modernization (BMDS) together with the Canadian government announced a transatlantic funding initiative totaling 300 million Canadian dollars. Germany is contributing approximately 100 million euros to this alliance, focusing financial resources on foundational research for provably safe systems.
The centerpiece of the funding package is LawZero, a non-profit research organization established by Turing Award winner Yoshua Bengio. Bengio, widely recognized as one of the founding figures of deep learning, has repeatedly raised concerns regarding the systemic risks posed by unchecked, goal-directed autonomous software agents. Under the agreement, LawZero will open an official research branch in Berlin, embedding cutting-edge safety research directly into the European academic and technological ecosystem.
The primary development project funded under this partnership is named Scientist AI. Designed as a mathematically rigorous model architecture, it strictly follows the principle of safety by design. Scientist AI relies on fact-grounded reasoning and formal verification without developing instrumental sub-goals or autonomous agency. This initiative positions itself as an open, public-interest alternative to the commercial, opaque autonomous agent platforms currently emerging from leading technology monopolies.
The international alliance is accompanied by domestic industrial policy measures aimed at digital independence. On September 17, 2026, the BMDS officially published the guidelines for its new funding program, Digital Tech-to-Product. This initiative provides up to 5 million euros per project to small and medium-sized enterprises as well as start-ups. The goal is to accelerate the transfer of foundational European digital technologies from research laboratories into mature, deployable market products.
The combined push for verified research and industrial commercialization reflects growing skepticism among European policymakers toward autonomous agentic workflows. While major software vendors increasingly grant autonomous execution capabilities to language models, Berlin and Ottawa are prioritizing formal boundaries and mathematical verification. For corporate leaders and IT strategists, this indicates an impending regulatory preference for transparent, bounded systems over open-ended autonomous agents.