CERTAIN: A European Competence Center for Trusted AI

The widespread and secure use of Artificial Intelligence (AI) in business and society requires reliable guarantees for fairness, transparency, and security. The CERTAIN Center at DFKI develops methods and standards for Trusted AI, enabling robust verification of the reliability and traceability of this technology.

CERTAIN is more than just a research project – it is a pioneering center for the development and certification of AI systems with strong guarantees. The long-term goal is to establish a leading European competence center for highly reliable AI technologies that closely connects research, industry, and society.

DFKI pursues these objectives by developing various methods to make the reliability of AI models demonstrable. The focus is on four main strategies:

• Guarantees by Design: AI systems are developed from the outset with verifiable safety and fairness features.
• Guarantees by Tools: Systematic testing methods ensure the reliability and robustness of AI.
• Guarantees by Insight: Simulations and visualizations provide transparency into critical decision-making processes.
• Guarantees by Interaction: Human feedback is integrated into the development process to optimize AI systems.

These core technical goals of CERTAIN aim to make AI models not only more efficient but also more comprehensible and reliable. Special attention is given to safety-critical applications.

Additionally, CERTAIN supports the vision of a European AI center modeled after CERN. The goal is to unite research, industry, and policy to establish European standards for Trusted AI and position Europe as a leading hub for secure AI development. A key aspect of this effort is close collaboration with standardization and certification organizations such as VDE and TÜV, as well as initiatives like Mission KI.

Presentation at Hannover Messe

At Hannover Messe, CERTAIN will showcase concrete examples of how these technical goals can be achieved. The demonstrations will include:

• Content checkers can pre-screen data used to train AI systems to ensure compliance with security and fairness requirements (Guarantees by Design).
• Systematic testing methods can verify the admissibility of outputs from generative AI systems (Guarantees by Tools).
• Visualization and explainability methods provide transparency into AI decision-making processes (Guarantees by Insight).
• Optimizations can be made through an interface, allowing new model predictions to be calculated with modified parameters (Guarantees by Interaction).

Beyond these technical presentations, CERTAIN will also highlight the political dimension of Trusted AI at the exhibition. This includes discussions on the proposal for a "CERN for AI" and the need for European standards and certification procedures. By combining technical demonstrations with strategic visions, CERTAIN aims to contribute to the establishment of Trusted AI on a European level while bridging the gap between research and industrial application.