Bitkom on the AI Security Institute

By establishing a security institute for artificial intelligence, Germany aims to catch up with countries such as the United Kingdom, the United States, and France. „These countries have already set up their own institutes,“ said Susanne Dehmel, a member of Bitkom’s executive board. „AI systems open up new possibilities, but they also bring new risks. By establishing a security institute for AI, Germany is laying the groundwork for being able to assess such risks thoroughly and at an early stage. The DE-AISI should be given a research mandate distinct from that of existing institutions and should compile a situation assessment with a focus on the new so-called frontier models. The focus must be on new systemic risks to Germany’s security and sovereignty. Issues related to labor, consumer, and data protection, as well as AI ethics, are already being competently addressed elsewhere. The new security institute will only be successful if we can attract top international talent. To achieve this, the institute needs funding at least on par with that of the British AISI.“

Germany's Path to World-Class Status

The capabilities of advanced AI systems are evolving at a rapid pace, and with them, their significance for national security, economic resilience, and technological sovereignty. Whether it’s cyberattacks, disinformation, or the misuse of powerful AI models, governments worldwide are increasingly investing in their own capabilities to better understand and assess the opportunities and risks of frontier AI. As part of the 2024 Seoul Declaration, Germany has committed to supporting the establishment of such institutions.

But what exactly should a German AI Security Institute look like? Which tasks should it take on, and which should it avoid? How can it successfully attract top international talent while avoiding duplicate structures with existing government agencies and research institutions? Bitkom has developed a concept for a world-class German AI Security Institute.

„With the establishment of a security institute for AI
Germany is creating the conditions to mitigate risks
”to be able to make a well-founded assessment at an early stage."

Susanne Dehmel,
Member of the Executive Board,
Bitkom

Security for Frontier AI

The focus is on the technical evaluation of advanced AI models, the preparation of security-related situation assessments for the federal government, and research into security measures for Frontier AI.

Key to success in this regard would be a clearly defined mandate without regulatory functions, an efficient technical infrastructure, international networking, and the ability to attract world-leading experts to Germany.

The rapid advances in the field of artificial intelligence have a direct impact on Germany’s national security and sovereignty. Not least, the novel cyber capabilities of Anthropic’s „Mythos“ AI model demonstrate this once again in a striking way. However, Germany is not adequately prepared for this—in part because the federal government lacks a sufficiently nuanced understanding of the capabilities and implications of frontier AI models. With the 2024 Seoul Declaration, Germany has explicitly committed to supporting the establishment and expansion of AI safety/security institutes and comparable institutions. The federal government now plans to fulfill this commitment, but critical questions regarding the development of its own government capabilities remain unresolved.

Bitkom Recommendations for Action

Germany should establish its own world-class AI Security Institute in the near term, modeled after the UK AI Security Institute (UK AISI). In its position paper, Bitkom identifies the following design aspects as key to the institute’s success:

The DE-AISI should be given a mandate exclusively for research and situation assessment regarding the capabilities, security measures, and impacts of frontier AI models. It should explicitly not be given a regulatory mandate, and its research mandate must be clearly distinguished from that of existing research institutions. The DE-AISI should focus thematically on the systemic impacts of frontier AI models on Germany’s national security and sovereignty—in particular on offensive cyber capabilities, misuse for the development of weapons of mass destruction, risks of loss of control, and mass manipulation. Other bodies would be responsible for data protection and AI ethics. In addition, adequate funding is needed, with an initial budget of at least 100 million euros for the first two years, as well as long-term follow-on funding of at least 75 million euros annually. (Source: Bitkom)