With the support of a newly established institute, the German government wants to better assess the opportunities and risks of advanced AI models. After the National Security Council decided in June 2026 to set up such an institute, AISI Germany, short for AI Safety and Security Institute, began its work in Berlin on August 31, 2026.
BSI and the Federal Network Agency form the core
The institute is being built up in stages. Initially, Germany's Federal Office for Information Security (BSI) and the Federal Network Agency (BNetzA) are to form a "virtual" core of the AI safety institute in a first phase. That means both the Digital Ministry and the Interior Ministry are responsible for AISI. The institute is modeled on international examples but will be shaped independently, based on Germany's specific needs.
It's meant to enable the government to identify safety-relevant risks early, derive evidence-based policy options, strengthen Germany's digital sovereignty in the age of AI, and strategically assess international developments. It's explicitly not a regulatory authority, but rather an analytical and operational center of expertise at the intersection of security, technology analysis, innovation, and strategic advice.
Concerns about loss of control over capable models
Alongside the technical evaluation of capable AI models, analyzing the safety-relevant effects of AI is among the institute's tasks. It's also meant to strategically advise the federal government on AI issues and build up national expertise and cooperation capacity in AI safety. When announcing AISI's launch, the Federal Digital Ministry pointed to recent publications on current AI safety incidents and to increasingly concrete risks from loss of control. Very capable AI models could therefore pose a threat to our digital environment and to the resilience of the economy, society, and democracy, the announcement stated.
Cooperation with European and transatlantic partners is meant to carry particular weight. According to the ministry, technical exchange is already underway between the participating agencies and with international partner institutions, including France, the United Kingdom, and the EU AI Office.
Open questions on budget and staffing
The announcement doesn't say how many positions or how much budget are planned for the institute; a corresponding inquiry to the two responsible ministries has so far gone unanswered. In a position paper, the German IT industry association Bitkom called for a German AI safety institute to be equipped with an initial budget of at least 100 million euros for the first two years, plus long-term secured follow-up funding of at least 75 million euros annually. Whether the actual funding will come close to these demands remains unclear for now.