Public Money, Public Code: Federal government releases AI modules

German administration is considered the epitome of sluggish processes and thick stacks of files. But when it comes to AI, the federal government wants to show that things can be done differently. The Federal Ministry for Digital and State Modernization (BMDS) has released a series of AI modules with the Spark project that could fundamentally change the way authorities work.

A special feature is the publication approach: under the guiding principle Public Money, Public Code, the applications are available to everyone on the OpenCode platform. In doing so, the BMDS follows the call for digital sovereignty and enables municipalities, companies, and civil society to use and further develop the tools without license fees.

Spark is intended to accelerate complex planning and approval procedures, which previously took months or even years, through intelligent operational assistance. Funding is provided by the Climate and Transformation Fund. The project aims to free up employees in approval authorities from routine tasks without replacing people. Rather, the AI prepares the flood of information from application documents in such a way that case workers can come to a well-founded decision more quickly. The final decision rests with humans.

Document Review Starting Point

Technically, Spark starts where it has been the biggest bottleneck so far: in reviewing and checking documents. The published modules cover typical tasks such as compiling relevant data from applications or performing a formal check for completeness and plausibility. The system is intended to detect when documents are missing or information is contradictory.

The core is a legal dogmatics supported by AI agents. It is directly connected to legal databases and can deconstruct and legally evaluate norms automatically. In the future, additional modules are to follow, which will also deal with substantive review and decision preparation.

Digital Minister Karsten Wildberger (CDU) sees the republic in a leading global role in AI-based administrative applications through this initiative. In February, the project won the award for the best use of AI in public services at the World Government Summit in Dubai. The open provision of source code is now intended to ensure that creative minds can adapt the modules and optimize them for various federal requirements. To boost this process, the BMDS is planning a two-day hackathon for June.

Security and Technical Hurdles

Despite the euphoria, the BMDS urges caution during implementation. The accompanying documentation contains a security note: the published code is to be understood as a reference and integration basis, but does not contain a ready-made security configuration for direct production use. Operators must harden the applications in their respective IT environments, define access rights, and manage sensitive data securely. A dedicated security review is mandatory before the AI is actually integrated into productive approval processes.

However, the entry barrier for developers is intended to be low. The modules are based on a Docker environment and can be set up quickly. Configuration is done via scripts. Interfaces to OpenAI-compatible endpoints and local solutions such as LiteLLM are provided.

(dahe)