The new research project “Human-AI Collaboration for Cybersecurity” (HAICC), conducted by the National Research Center for Applied Cybersecurity ATHENE, addresses precisely this challenge. It investigates how collaboration between humans and AI in solving complex and underexplored tasks in cybersecurity can be specifically improved. The collaborative project is coordinated by Professor Iryna Gurevych, holder of the first ATHENE Distinguished Professorship, at the Technische Universität Darmstadt and will run from 2026 to 2029.

At the core of the project is a collaborative approach: rather than relying on fully autonomous systems, human experts and AI agents work together to solve complex security problems. The collaboration is structured as an iterative process in which humans and AI agents tackle complex problems together, as equal partners, quickly and effectively.

Powerful, but also transparent and controllable

A particular focus lies on the integration of contextual and human preferences, the use of multimodal data, and natural language interaction. AI systems should not only be powerful, but also transparent and controllable, so that experts can review and contextualize the results they produce. With this approach to designing collaboration, the project addresses key challenges of generative AI such as the so-called automation bias, that is, the uncritical acceptance of AI-generated recommendations, as well as the limited availability of data in cybersecurity.

The approaches developed in the project will be tested in scenarios such as the analysis of vulnerabilities and security protocols. AI agents help structure and process large and heterogeneous data sets, while human experts remain responsible for evaluation, interpretation, and further development. The aim is to increase efficiency without compromising the quality and reliability of decisions.

Several research groups at TU Darmstadt are involved in the project. These include the Artificial Intelligence & Machine Learning Lab, the Data and AI Systems Lab, the Multimodal AI Lab, PEASEC – Science and Technology for Peace and Security, the Security in Information Technology Group, the Software Technology Group, the Ubiquitous Knowledge Processing (UKP) Lab, the Urban Interaction Lab, and the Visual Inference Lab. The Chair of Cybersecurity at Goethe University Frankfurt, a partner of TU Darmstadt in the Rhine-Main Universities Alliance (RMU), which also includes Johannes Gutenberg University Mainz, is also involved. Ludwig Maximilian University of Munich receives a subcontract and contributes its dedicated expertise on the dynamic memory of AI agents.