Germany has launched a new authority for the safety of Artificial Intelligence with AISI Deutschland. The institute began its work on August 31, 2026, in Berlin and consolidates the competencies of the Federal Office for Information Security and the Federal Network Agency. It is intended to examine AI models before they pose risks to users or infrastructure.
Federal Office for Security and Network Agency Share Responsibilities
The name AISI Deutschland stands for AI Security and Safety Institute Germany. The institute initially brings together two federal agencies with different mandates: The Federal Office for Information Security takes on the security part, meaning the protection of AI systems from targeted attacks, data manipulation, and attempts at misuse. The Federal Network Agency is responsible for safety, ensuring that AI systems do not create unintended dangers for users or the environment and function reliably within defined parameters.
BSI President Claudia Plattner classifies Artificial Intelligence as a central security challenge according to a report from the Westdeutsche Zeitung. AISI Deutschland explicitly does not see itself as a traditional supervisory authority with sanctioning rights, but as an analytical-operational competence center, a so-called think-and-do tank. It is to assess the opportunities and risks of new AI models for Germany, develop solutions for the safe handling of the systems, and advise the federal government, administration, economy, and civil society.
Additionally, the institute is to work with European and transatlantic partners on uniform standards for particularly powerful models. The institution is based in Berlin, and the federal government has not yet provided information on personnel strength and budget.
National Security Council Decides on Establishment in June
The foundation for the establishment is a resolution of the National Security Council from June 2026. According to a statement from the federal government, the institute provides the basis for Germany to keep pace with the international development of Artificial Intelligence. It is being established and led jointly by the Federal Ministry of Digital and the Federal Ministry of the Interior.
Federal Interior Minister Alexander Dobrindt points out that AI must remain controllable while the state must also participate in its rapid development. Digital Minister Karsten Wildberger had only recently formulated a new AI interim goal for Germany at the cabinet retreat in Neuhardenberg in late August.
The digital association Bitkom welcomed the establishment of the institute back in June, but called for an independent research mandate focusing on systemic risks of so-called frontier models instead of duplicating existing roles for labor, consumer, or data protection. None of the statements available so far mention a connection with the reporting and testing obligations of the EU AI Regulation, leaving coordination with the Brussels AI oversight open. There are also no details available yet on a fixed testing calendar or the first models to be evaluated.
International Models Provide Blueprint and Warning
With its establishment, Germany is catching up to countries like the UK, the USA, and France, which have already set up their own AI security institutes. According to Bitkom board member Susanne Dehmel, the German counterpart needs at least funding at the level of the British AI Security Institute to be competitive in the international race for skilled workers.
The British counterpart is already publishing its own analyses of AI models, such as the cyber lag of open language models compared to proprietary systems – a task category that will likely also fall to AISI Deutschland in the future. It is precisely such an examination before market readiness that Bitkom has in mind with its demand for an independent research mandate.
How unstable such institutions can be is shown by the American counterpart CAISI: There, the responsible head resigned after only three months in office, already the third AI official of the US government within a year. For AISI Deutschland, it remains open how stable its leadership structure between the involved ministries and agencies will be in the long term. No concrete testing reports from the new German institute are available so far.
It will be crucial whether AISI Deutschland can actually examine new models before market readiness or whether its assessments, as with some foreign counterparts, will only take effect after incidents. The funding also remains open: Without a budget that is competitive from Bitkom’s perspective, the young institute may find it difficult to attract experts from better-equipped institutions in London or Washington.
