German AI Safety Institute Comes to Berlin: Germany Follows Suit in AI Safety
Germany is responding to the rapid development of Artificial Intelligence and incidents of AI agents breaking out of test environments by establishing the German AI Safety Institute (AISI Germany). This new institute, based in Berlin, will be responsible for evaluating risky AI systems.
The facility aims to systematically assess the opportunities and dangers of advanced AI models and to sustainably strengthen the country’s digital resilience. In its initial phase, AISI Germany will operate as a virtual nucleus, with two established federal agencies sharing the core tasks: The Federal Office for Information Security (BSI) will cover IT security, while the Federal Network Agency will be responsible for classic safety issues, primarily physical security and product suitability.
Federal Digital Minister Karsten Wildberger (CDU) emphasized the critical importance of technically sound and rapid evaluation of state-of-the-art AI systems for Germany as a business location. Federal Interior Minister Alexander Dobrindt (CSU) stressed that while technology offers opportunities for society and the economy, “AI must remain controllable.”
Focus on IT Risks and Loss of Control
The content focus will initially be on a thorough analysis of leading models, particularly concerning IT risks and potential loss of control. Recent incidents have highlighted how dangerous powerful AI models can be for democracy, the economy, and civil society. In a second phase, the institute will be expanded in terms of personnel and thematic scope to serve as a central knowledge hub and advisor for politics, administration, and business.
Location Debate and International Precedent
The choice of Berlin as the main location has not been universally welcomed by experts. Within the IT and scientific communities, there has long been an intense debate about whether concentrating solely on the capital does justice to the complexity of Germany’s research landscape. Critics argue that AI centers and computer science faculties in Southern Germany or North Rhine-Westphalia would also be suitable homes for such an institute. From this perspective, greater decentralization would have allowed for deeper integration with established top research institutions.
AISI Germany is based on a resolution by the national security council from June and is modeled after the British AI Safety Institute. The latter has pioneered state-led scrutiny of so-called frontier models. Similar to its British counterpart, the German institution will also focus on global networking from the outset, close cooperation with European partners, and exchange with the EU AI Office and bodies in the USA or France.
Further Demands from Academia
Further demands are coming from the scientific community. AI experts like Kristian Kersting, a professor at TU Darmstadt, warn that state security institutes can only be one component. To achieve technical sovereignty, Germany needs more investment in its own infrastructure and developments, open models like Open LLMs, or specialized initiatives such as the Soofi project.
