Germany Launches AI Security Institute to Test Advanced Models

Germany just took a major step toward making advanced AI safer by launching its own national AI Security Institute.

The country's National Security Council approved the German AI Security Institute (DE-AISI) in June 2026, and by August, experts from two federal agencies were already collaborating on the project. The Federal Office for Information Security and Federal Network Agency are leading the initial work while the government figures out the permanent setup.

The institute's main job is giving Germany's government the technical know-how to assess advanced AI models and spot potential security risks. Right now, the focus is heavily on cybersecurity threats that sophisticated AI systems might create or enable.

But the mission could grow much wider. Proposals being considered would expand the institute's work to include evaluating risks from chemical, biological, radiological and nuclear threats, plus concerns about AI systems that might become difficult to control.

Germany isn't building this alone. The team is already working with partner institutes in France and the UK, along with the EU AI Office, sharing technical knowledge and best practices.

The Ripple Effect

This institute could strengthen AI safety far beyond Germany's borders. By joining the international network of AI Safety and Security Institutes, Germany is helping build a global system for evaluating advanced AI before problems emerge.

The institute is designed to complement the EU AI Act, not duplicate it. DE-AISI will provide scientific expertise and risk assessments, while existing agencies keep their regulatory powers.

The government still needs to make some big decisions. Four cities are competing to host the permanent facility: Berlin, Bonn, Munich, and Saarbrücken. Berlin might win if advising federal ministries becomes central to the work, since it's close to government offices and has a strong AI talent pool.

Money and staffing remain open questions too. Germany's Digital Minister wants "world class experts," but AI specialists often earn salaries well above typical government pay. One solution being considered is setting up the institute as a federally owned company, similar to Germany's innovation agency SPRIND, which would allow more flexibility on salaries and spending.

Proposed budgets range from €60 million to at least €75 million annually, roughly matching the UK AI Security Institute's £66 million budget.

By building expertise now, Germany is preparing to handle AI challenges before they become crises.