Missing Link: Invisible Revolution – how the Federal Government is Flooding the Administration with AI

Summary

The Federal Government can no longer clearly show where Artificial Intelligence is being used throughout the administration. AI has evolved from an exotic project into an everyday component in standard software and is diffusing through all levels of German bureaucracy. The Federal Ministry for Digital Affairs and State Modernization (BMDS) admits that a "clear demarcation" is no longer possible – a fundamental shift compared to July 2024, when over 200 AI applications were still being discussed. The response to a parliamentary inquiry from the Left Party faction reveals both opportunities in innovative projects and significant control deficits in security-sensitive areas.

People

Topics

Detailed Summary

The New Lack of Transparency

The Federal Government's digital agenda has reached a turning point. While the government still enumerated a manageable amount of over 200 AI applications in 2024 with an impenetrable classified section, it now capitulates before the sheer volume. The BMDS admits that AI is already a component of firewalls, word processors, and standard office software. The technology has evolved from a "flagship project" into an everyday "light bulb" – present everywhere and effectively rendered invisible.

Critical Questions from the Left

The Left Party faction in the Bundestag expresses deep skepticism toward AI deployment in "fundamental rights-sensitive areas." The faction warns of:

• Discrimination Potential: Algorithmic bias can lead to disadvantage for women and people with a migration background
• Security Powers: Plans by the black-red coalition to grant security authorities extensive powers for automated data analysis and to train AI with real data
• Energy Consumption: The massive electricity consumption for complex models is barely considered in public debate; the Left Party faction demands mandatory CO₂ footprint disclosures as a procurement criterion

The Silence of the Services

The government indirectly confirms that AI is already a tool of security policy – through its silence. The Federal Intelligence Service (BND), the Federal Office for the Protection of the Constitution (BfV), and the Military Counterintelligence Service (MAD) refuse information across the board, and even confidential responses are reportedly not possible.

The justification reveals the strategic dimension: If AI methods like "text recognition" combined with specific data sources became known, adversaries could draw conclusions about technical capabilities. Worse still – if AI training databases became public, adversaries could deliberately "poison" them (data poisoning) to manipulate the systems. The Federal Ministry of Defense also remains tight-lipped and points to possible conclusions about troop combat power.

Critics argue that precisely where fundamental rights are most at risk, parliamentary oversight of AI deployment is lacking.

From Patchwork to Central Platform

Compared to the situation in early 2024, a strategic shift is evident: Rather than each department developing its own chatbot, the federal government is now moving toward centralization and platform economics.

MaKI is the new central transparency register – a "matching platform" where agencies can see what others have already developed. Since November 2024, federal states and municipalities also have access to bridge the federal patchwork technologically.

Kipitz is the operational centerpiece: a planned AI platform for the federal administration, operated by the ITZBund Service Center. It aims to solve the "ChatGPT dilemma" of the administration by providing generative AI models through secure interfaces – without sensitive government data landing on servers of US tech giants. Kipitz is a "closed-source proprietary development" using open-source models. For 2026, 1.7 million euros in budget funds are planned, with hardware costs totaling around 40 million euros.

A Fraunhofer analysis recommends federal agencies use interchangeable open-source-based LLM (Large Language Model) solutions – such as Meta's Llama, Google Gemma, or DeepSeek. While this strengthens digital sovereignty, it leaves a strategic gap: given evolving open-source understanding, researchers are toying with the idea of developing their own European LLM.

Innovative Applications

Beyond text summarization, there are a number of remarkable projects:

• BIKO-UA: Image recognition software for identifying victims of war in Ukraine – humanitarian and forensic application
• BAMF: Deployment of AI models for better assessment of migration movements
• KIResQ: AI-supported evaluation of thermal images for faster discovery of missing persons in rough terrain
• Silva: AI-controlled drones for automated search for wildfire sources from the air
• Federal Institute for Hydrology (BfG): AI deployment for detecting plastic in rivers and oil slicks on the sea
• German Meteorological Service (DWD): Planned "AI center" for improved weather forecasting and "nowcasting" (short-term forecasting) to protect against extreme weather events
• Climate Protection: AI for examining rock formations and predicting groundwater levels
• FACTSBot: System for detecting and validating machine-generated content
• Nebula: User-centered system for detecting fake news
• SpeechTrust+: Tool for detecting AI-based speech synthesis and voice distortion – against "grandchild scam 2.0" and political manipulation

Key Findings

• The Federal Government is losing sight of AI deployment in the administration because the technology is already integrated into standard software and no longer acts in isolation.

• In security-sensitive areas (BND, BfV, MAD), the government refuses information completely – here the "administrative revolution" ends and reasons of state begin.

• The Left Party faction criticizes missing protection mechanisms against discrimination through algorithmic bias and demands transparency on CO₂ footprints.

• Strategic shift from decentralized "patchwork" to centralized platforms (MaKI, Kipitz) for better control and digital sovereignty.

• Kipitz platform with 40 million euros in hardware costs is intended to prevent sensitive government data from landing on US servers – focus on open-source models like Llama, Gemma, and DeepSeek.

• Beyond administrative efficiency, innovative AI applications are emerging in humanitarian, security, and climate-relevant areas (BIKO-UA, Silva, nowcasting, disinformation).

• Fraunhofer analysis recommends European open-source strategy but is also considering development of a standalone European large language model to reduce strategic dependencies.

Metadata