Gustav Keppler

Gustav Keppler, Ghada Elbez, Veit Hagenmeyer

The rapid evolution and use of Large Language Models (LLMs) in professional workflows require an evaluation of their domain-specific knowledge against industry standards. We introduceCyberCertBench, a new suite of Multiple Choice Question Answering (MCQA) benchmarks derived from industry recognized certifications. CyberCertBench evaluates LLM domain knowledgeagainst the professional standards of Information Technology cybersecurity and more specializedareas such as Operational Technology and related cybersecurity standards. Concurrently, we propose and validate a novel Proposer-Verifier framework, a methodology to generate interpretable,natural language explanations for model performance. Our evaluation shows that frontier modelsachieve human expert level in general networking and IT security knowledge. However, theiraccuracy declines in questions that require vendor-specific nuances or knowledge in formalstandards, like, e.g., IEC 62443. Analysis of model scaling trend and release date demonstratesremarkable gains in parameter efficiency, while recent larger models show diminishing returns.Code and evaluation scripts are available at: https://github.com/GKeppler/CyberCertBench.

Moritz Gstür, Gustav Keppler, Mohammed Ramadan et al.

Critical energy infrastructures increasingly rely on information and communication technology for monitoring and control, which leads to new challenges with regard to cybersecurity. Recent advancements in this domain, including attribute-based access control (ABAC), have not been sufficiently addressed by established standards such as IEC 61850 and IEC 62351. To address this issue, we propose a novel real-time server-aided attribute-based authorization and access control for time-critical applications called RTS-ABAC. We tailor RTS-ABAC to the strict timing constraints inherent to the protocols employed in substation automation systems (SAS). We extend the concept of conventional ABAC by introducing real-time attributes and time-dependent policy evaluation and enforcement. To safeguard the authenticity, integrity, and non-repudiation of SAS communication and protect an SAS against domain-typical adversarial attacks, RTS-ABAC employs mandatory authentication, authorization, and access control for any type of SAS communication using a bump-in-the-wire (BITW) approach. To evaluate RTS-ABAC, we conduct a testbed-based performance analysis and a laboratory-based demonstration of applicability. We demonstrate the applicability using intelligent electronic devices, merging units, and I/O boxes communicating via the GOOSE and SV protocol. The results show that RTS-ABAC is able to secure low-latency communication between SAS devices, as up to 99.82 % of exchanged packets achieve a round-trip time below 6 ms. Moreover, the results of the evaluation indicate that RTS-ABAC is a viable solution to enhance the cybersecurity not only in a newly constructed SAS but also via retrofitting of existing substations.

Gustav Keppler, Moritz Gstür, Veit Hagenmeyer

The advancement of Large Language Models (LLMs) has raised concerns regarding their dual-use potential in cybersecurity. Existing evaluation frameworks overwhelmingly focus on Information Technology (IT) environments, failing to capture the constraints, and specialized protocols of Operational Technology (OT). To address this gap, we introduce CritBench, a novel framework designed to evaluate the cybersecurity capabilities of LLM agents within IEC 61850 Digital Substation environments. We assess five state-of-the-art models, including OpenAI's GPT-5 suite and open-weight models, across a corpus of 81 domain-specific tasks spanning static configuration analysis, network traffic reconnaissance, and live virtual machine interaction. To facilitate industrial protocol interaction, we develop a domain-specific tool scaffold. Our empirical results show that agents reliably execute static structured-file analysis and single-tool network enumeration, but their performance degrades on dynamic tasks. Despite demonstrating explicit, internalized knowledge of the IEC 61850 standards terminology, current models struggle with the persistent sequential reasoning and state tracking required to manipulate live systems without specialized tools. Equipping agents with our domain-specific tool scaffold significantly mitigates this operational bottleneck. Code and evaluation scripts are available at: https://github.com/GKeppler/CritBench