Research at KASTEL SRL

Secure Federated Machine Learning

Machine learning models require large amounts of data to be trained on, leaving little training data available in use cases with strong privacy requirements.The Secure Federated Machine Learning project of SAP and KASTEL Security Research Labs is investigating approaches for training a neural network on data from multiple parties without having to reveal their data to the other parties. These approaches can then be used in use cases with strong privacy requirements.

Institute of Science Tokyo, University of Malaga: Trustworthy Self-adapting Systems

This research addresses the challenge of providing trustworthy assurance for software intensive systems operating under multiple, interdependent uncertainties. By explicitly capturing and analyzing how different uncertainty types interact and produce emergent effects, we seek to improve safety/performance guarantees and strengthen stakeholder trust in autonomy for critical tasks.

• Organization of the NII Shonan Meeting, Seminar No. 232 “Uncertainty Interaction in Software-intensive Systems (UNISON)”, March 2–5, 2026, Shonan Village, Japan. Seminr report: to appear.
• D. Weyns, R. Calinescu, R. Mirandola, K. Tei, M. Acosta, A. Bennaceur, N. Boltz, T. Bures, J. Cámara, A. Diaconescu, G. Engels, S. Gerasimou, I. Gerostathopoulos, S.G. Yaman, V. Grassi, S. Hahner, E. Letier, M. Litoiu, L. Marsso, A. Musil, J. Musil, G.N. Rodrigues, D. Perez-Palacin, F. Quin, P. Scandurra, A. Vallecillo & A. Zisman (2023): Towards a Research Agenda for Understanding and Managing Uncertainty in Self-Adaptive Systems. – ACM SIGSOFT Software Engineering Notes, 48 (4): 20–36.

Singapore University of Technology and Design: Cyber Secure Critical Infrastructure: Joint Lab Experiments KASTEL Security Lab Energy & iTrust

The collaboration will investigate methods for securing critical infrastructure against cyberattacks using real-world lab environments. KIT uses the KASTEL Security Lab Energy, while the Singapore University of Technology and Design contributes the iTrust Labs.

Stanford University, Computer Science Department and Law School: Operationalizing ‘Risk’ as a Legal and Computational Concept under the EU AI Act

The EU AI Act uses a risk-based approach to regulate AI systems, calibrating the intensity of regulation according to the risks they pose. While the term ‘risk’ implies quantification, resulting from the combination of the probability and severity of harm, the AI Act refers to risks to fundamental rights, thereby engaging a qualitative perspective. In this piece, we address this puzzle using a framework that integrates risks with the protection of fundamental rights, the legitimate purposes of providers and deployers, and the cost of regulatory measures. We discuss this framework against the backdrop of potential approaches to quantifying risks, with a specific focus on defining and measuring the main elements of the concept of risk: probability, severity and their combination. We conclude that the approach to quantifying risks directly impacts the classification of AI systems under the AI Act and, consequently, the applicable rules. If the quantification of risks is left to providers and deployers, however, there is potential for ‘risk hacking’, which could lead to the underclassification of AI systems and regulatory shortcuts.

Attack Detection in Heterogenous Systems

The joint project with ABB Corporate Research Center in Mannheim investigates the security and resilience of heterogeneous system architectures against external manipulation. It involves the development and research of methods for both detecting and analyzing the causes of attacks on learning systems.

IIP 2.0 – Certification of Privacy-Enhancing Technologies

KIT was contacted by S.A.F.E. e.V. (Software Alliance for E-mobility), an association of manufacturers of electronic vehicle charging stations, because a patent troll managed to obtain a patent on the de-facto standard cryptographic scheme used for the billing process of charging stations. The researchers at KIT developed a solution, called “IIP 2.0”, that circumvented the patent, provided an expert opinion and quantitative security proof. Based on this expert opinion and the reference implementation by S.A.F.E. e.V., “IIP 2.0” was officially certified by the National Metrology Institute (Physikalisch-Technische Bundesanstalt, PTB) and rolled out. With the existence of a viable alternative, the patent troll eventually allowed to use the de-facto standard free of charge.

Sec4IoMT: Security for the Internet of Medical Things

The goal of the BMFTR-funded “Security for the Medical Internet of Things” (Sec4IoMT) project is to develop together with Infineon Technologies AG and WIBU SYSTEMS AG economically viable and technically secure solutions for the long-term protection of medical devices. The basis for this is secure communication channels. For this reason, corresponding security architectures and security retrofitting processes—that is, processes in which existing devices are adapted to current security requirements—are being developed, tested by medical device manufacturers for practical suitability, and analyzed for their cost-effectiveness. Research is being conducted on how updates to software, firmware, libraries of cryptographic algorithms, certificates, and keys can be installed and distributed even for large quantities and diverse users without jeopardizing product approval. To this end, the devices are to remain with the user and be online only temporarily.

Cryptographic Enclave

Researchers at KIT have developed the “cryptographic enclave”, a solution for secure, legally compliant, high-performance and scalable computations on sensitive data. The “cryptographic enclave” is based on a sophisticated security architecture in combination with cryptography. KIT and SAP are working together to incorporate techniques from the cryptographic enclave into the product “SAP Sovereign Cloud”. Together with TÜV Nord, SAP is in the process of writing the inspection catalogue for data centers of the German federal government. Lessons learned from the cryptographic enclave will be incorporated to further improve security. In order to enable this transfer, KIT and SAP work on realizing an advanced use-case provided by SAP using the cryptographic enclave. Research on the cryptographic enclave was partially funded by a three-year grant for the project “Secure Federated ML”.