SOVEREIGN

SOVEREIGN presenting their work at the Milestone Symposium at the Cyberagentur

Tue, 05 May 2026 17:00:00 +0200

From April 27th to 29th, 2026, we had the opportunity to participate in the Milestone Symposium organized by the Cyberagentur (Agency for Innovation in Cybersecurity). The event provided an excellent platform for exchange between German government agencies, the research projects they support, and the wider cybersecurity research community.

We contributed with multiple presentations as well as poster sessions, and greatly valued the discussions around current solutions and the future direction of cybersecurity research in securing critical infrastructure.

You can find the list of published papers at https://sovereign-project.de/publications/.

Accepted Paper at SBFT Workshop at ICSE 2026

Tue, 10 Mar 2026 13:00:00 +0200

We are happy to announce that our paper “Coverage-Guided Multi-Agent Harness Generation for Java Library Fuzzing” is accepted and will appear at The 19th International Workshop on Search-Based and Fuzz Testing in Rio de Janerio during ICSE 2026.

Abstract: Coverage-Guided Multi-Agent Harness Generation for Java Library Fuzzing

Coverage-guided fuzzing has proven effective for software testing, but targeting library code requires specialized fuzz harnesses that translate fuzzer-generated inputs into valid API invocations. Manual harness creation is time-consuming and requires deep understanding of API semantics, initialization sequences, and exception handling contracts. We present a multi-agent architecture that automates fuzz harness generation for Java libraries through specialized LLM-powered agents. Five ReAct agents decompose the workflow into research, synthesis, compilation repair, coverage analysis, and refinement. Rather than preprocessing entire codebases, agents query documentation, source code, and callgraph information on demand through the Model Context Protocol, maintaining focused context while exploring complex dependencies. To enable effective refinement, we introduce method-targeted coverage that tracks coverage only during target method execution to isolate target behavior, and agent-guided termination that examines uncovered source code to distinguish productive refinement opportunities from diminishing returns. We evaluated our approach on seven target methods from six widely-deployed Java libraries totaling 115,000+ Maven dependents. Our generated harnesses achieve a median 26% improvement over OSS-Fuzz baselines and outperform Jazzer AutoFuzz by 5% in package-scope coverage. Generation costs average $3.20 and 10 minutes per harness, making the approach practical for continuous fuzzing workflows. During a 12-hour fuzzing campaign, our generated harnesses discovered 3 bugs in projects that are already integrated into OSS-Fuzz, demonstrating the effectiveness of the generated harnesses.

Accepted Paper at DeepTest Workshop at ICSE 2026

Tue, 10 Mar 2026 13:00:00 +0200

We are happy to announce that our paper “Beyond Accuracy: Characterizing Code Comprehension Capabilities in (Large) Language Models” is accepted and will appear at the Seventh International Workshop on Deep Learning for Testing and Testing for Deep Learning (DeepTest 2026) during ICSE 2026.

Abstract: Beyond Accuracy: Characterizing Code Comprehension Capabilities in (Large) Language Models

Large Language Models (LLMs) are increasingly integrated into software engineering workflows, yet current benchmarks provide only coarse performance summaries that obscure the diverse capabilities and limitations of these models. This paper investigates whether LLMs’ code-comprehension performance aligns with traditional human-centric software metrics or instead reflects distinct, non-human regularities. We introduce a diagnostic framework that reframes code understanding as a binary input-output consistency task, enabling the evaluation of classification and generative models. Using a large-scale dataset, we correlate model performance with traditional, human-centric complexity metrics, such as lexical size, control-flow complexity, and abstract syntax tree structure. Our analyses reveal minimal correlation between human-defined metrics and LLM success (AUROC 0.63), while shadow models achieve substantially higher predictive performance (AUROC 0.86), capturing complex, partially predictable patterns beyond traditional software measures. These findings suggest that LLM comprehension reflects model-specific regularities only partially accessible through either human-designed or learned features, emphasizing the need for benchmark methodologies that move beyond aggregate accuracy and toward instance-level diagnostics, while acknowledging fundamental limits in predicting correct outcomes.

Accepted Paper at AsiaCCS 2026

Wed, 28 Jan 2026 15:00:00 +0200

We are happy to announce that our paper “CoCoRADE: Introducing Replay Protection for Data-at-Rest of Confidential Virtual Machines” is accepted and will appear at the 21st ACM ASIA Conference on Computer and Communications Security (AsiaCCS'26).

Abstract: CoCoRADE: Introducing Replay Protection for Data-at-Rest of Confidential Virtual Machines

The cloud has become a popular tool for companies to host and process data. To keep this data safe from cloud providers, multiple Confidential Computing technologies have emerged. Confidential Virtual Machines protect the memory of VMs and thus all data in use. Full Disk Encryption (FDE) and Authenticated Disk Encryption (ADE) solutions are commonly used to store data at rest on a host-provided disk. However, disk encryption does not provide replay protection, meaning an older disk version or parts of an older disk can be replayed by the host to the Confidential VM. With no countermeasures taken, this is impossible to detect and opens new attack vectors, e.g., allowing to undo critical changes made to configuration files and data. In this paper, we solve this problem by proposing CoCoRADE, a way to store data on a disk while maintaining confidentiality, integrity and replay protection of data. In addition, we highlight how freshness is maintained across reboots. Our proof of concept demonstrates how this is easily added to existing Confidential VM offerings in the cloud without needing any additional support from the cloud provider or specialized hardware. The evaluation results show that CoCoRADE provides more security while performing comparably to, if not better than, existing ADE solutions with acceptable overheads in memory usage and remount time.

Internal Project Workshop at Deutsche Cyber-Sicherheitsorganisation (DCSO)

Thu, 20 Nov 2025 12:00:00 +0100

The SOVEREIGN project, funded by the Cyberagentur (Agentur für Innovation in der Cybersicherheit) and coordinated by the Universität Hamburg, held its fourth internal workshop at the DCSO in Berlin on November 13-14th, 2025.

The workshop disussed current and future directions of research of the consortium partners. Detailed discussions centered on the progress of individual work packages and the continuous efforts to achieve seamless integration of the components contributed by project members. These integration activities are vital for establishing a cohesive architectural framework, which is essential for developing a comprehensive understanding of cyber threats and is critical for progressing in the attribution of threat actors. The meeting concluded with numerous open research directions and ideas on how to leverage the collected attack information within a graph-based framework to support attribution.

SOVEREIGN presenting at Conference on Communications and Network Security (CNS'25)

Wed, 15 Oct 2025 16:00:00 +0200

The SOVEREIGN project, focusing on critical infrastructure protection, presented two research papers at the 2025 IEEE Conference on Communications and Network Security (CNS) in Avignon, France. Our first paper introduced a machine learning–based method for tagging security alerts with MITRE ATT&CK tactics, improving threat context and streamlining analyst workflows. The second paper proposed a real-time detection framework using kill chain state machines to identify multi-stage cyberattacks as they progressed. These contributions demonstrated SOVEREIGN’s commitment to sovereign, explainable, and operational cybersecurity solutions designed to strengthen proactive defense in critical digital environments.

Anum Talpur, Jannik Schröder, Liliana Kistenmacher, Wolfram Wingerath, Georg Becker, Mathias Fischer. "Tagging Alerts to Adversaries: ML-Enabled Classification Using MITRE ATT&CK Framework". IEEE Conference on Communications and Network Security. Avignon, France. 2025.

PDF BibTeX

Liliana Kistenmacher, Anum Talpur, Mathias Fischer. "Real-Time Detection of Multi-Stage Attacks using Kill Chain State Machines". IEEE Conference on Communications and Network Security. Avignon, France. 2025.

PDF BibTeX Source Code Reproducible Experiments

Internal Project Workshop at University of Lübeck

Fri, 11 Jul 2025 16:00:00 +0200

The SOVEREIGN project, funded by the Cyberagentur (Agentur für Innovation in der Cybersicherheit) and coordinated by the Universität Hamburg, held its third internal workshop at the Institute for IT-Security of the University of Lübeck on July 10-11th, 2025.

Hosted in a collaborative setting, the workshop brought together all project partners to assess current progress and align on key objectives. In-depth discussions focused on the status of individual work packages and the ongoing efforts to ensure the tight integration of project members’ components. These integration activities are crucial for enabling a holistic view of attacks and wave the path towards attribution of cyber threats. The meeting concluded with a clear definition of interfaces between project partners.

Accepted Paper at the Informatik Festival 2025 Workshop

Fri, 11 Jul 2025 15:00:00 +0200

We are happy to announce that our paper “From Logs to Language: Making OCSF the Open API for Security Data with Tenzir” has been accepted at the Informatik Festival 2025 Workshop for Cybersecurity: Prävention, Detektion und Reaktion mit Open Source-Perspektiven.

Accepted Papers at CNS'25

Thu, 10 Jul 2025 10:00:00 +0200

Double tap at the IEEE Conference on Communications and Network Security (CNS 2025)!

Abstract: Tagging Alerts to Adversaries: ML-Enabled Classification Using MITRE ATT&CK Framework

An intrusion detection system (IDS) in an enterprise network plays an important role in identifying and alerting suspicious activity. The triggered alerts are generated along with false alarms where a legitimate activity is mistakenly classified as malicious and results in alerts fatigue. Too many alerts that are not integrated and lack related context and correlation can burden security analytics and make prioritization of threats more challenging. For security operations centers (SOCs) in an enterprise, it can result in critical alerts being overlooked or undetected, which can lead to dire attacks not being identified on time. Therefore, an automated tagging of alerts for prioritization of high-risk events is important. To overcome this challenge, we use machine learning (ML) to enrich the alerts of IDS with the real-world adversary information of the MITRE ATT&CK framework. In particular, we leverage the knowledge of the MITRE ATT&CK matrix for enterprise and apply multi-layer perceptron (MLP) and transformer-based learning in a novel way to uncover the possible correlation of alerts to known adversarial behaviors (or tactics). We evaluate our models over the recent security logs of the real enterprise network and demonstrate an accuracy of up to 95% with our automated alert classification. Extensive experiments are performed with publicly available datasets as well to demonstrate the performance of the transformer model and verify its effectiveness against different IDS setups.

Abstract: Real-Time Detection of Multi-Stage Attacks using Kill Chain State Machines

Cyber threats present an ongoing challenge for organisations worldwide. Attackers range from cyber criminals to state-funded groups that have a specialised skill set to execute complex attacks and present an Advanced Persistent Threat(APT). Therefore, organisations use security monitoring as a second line of defence to detect attacks based on signatures that raise alarms when an Indicator of Compromise (IoC) is observed. However, current Intrusion Detection Systems (IDS) generate many false positives, leading to alert fatigue. The raised alerts also do not show the whole attack as they need to be dissected individually. Our work presents a simplified approach that enables efficient and real-time construction of attacks by correlating alerts. Tests with different network datasets suggest that our prioritisation mechanisms can reduce the number of false-positive alerts by 99%. Our performance evaluation indicates that we can detect multi-stage attacks in real time with a low memory footprint and short execution time.

SOVEREIGN presenting at SIC 2025 Symposium

Mon, 26 May 2025 18:00:00 +0200

From May 14th to 15th, 2025, we had the opportunity to present our work at the SIC 2025 Symposium in Halle (Saale), Germany. The event, hosted by the Cyberagentur (Agency for Innovation in Cybersecurity), served as a valuable platform for connecting German government agencies with the research projects they fund, as well as with the broader cybersecurity research community. We were excited to present our work and engage in discussion about the future of cyber security.

Accepted Paper at DSN 2025

Thu, 15 May 2025 00:00:00 +0200

Accepted Paper at ARES 2025

Thu, 15 May 2025 00:00:00 +0200

Accepted Paper at the ACNS

Thu, 08 May 2025 10:00:00 +0200

Our paper was accepted at the 23rd International Conference on Applied Cryptography and Network Security (ACNS'25).

Abstract:

In an era where cyberattacks increasingly target the software supply chain, the ability to accurately attribute code authorship in binary files is critical to improving cybersecurity measures. We propose OCEAN, a contrastive learning-based system for function-level authorship attribution. OCEAN is the first framework to explore code authorship attribution on compiled binaries in an open-world and extreme scenario, where two code samples from unknown authors are compared to determine if they are developed by the same author. To evaluate OCEAN, we introduce new realistic datasets: CONAN, to improve the performance of authorship attribution systems in real-world use cases, and SNOOPY, to increase the robustness of the evaluation of such systems. We use CONAN to train our model and evaluate on SNOOPY, a fully unseen dataset, resulting in an AUROC score of 0.86 even when using high compiler optimizations. We further show that CONAN improves performance by 7% compared to the previously used Google Code Jam dataset. Additionally, OCEAN outperforms previous methods in their settings, achieving a 10% improvement over state-of-the-art SCS-Gan in scenarios analyzing source code. Furthermore, OCEAN can detect code injections from an unknown author in a software update, underscoring its value for securing software supply chains.

Internal Project Workshop at Fraunhofer AISEC

Tue, 01 Apr 2025 12:00:00 +0100

The SOVEREIGN project, funded by the Cyberagentur (der Agentur für Innovation in der Cybersicherheit) and led by the Universität Hamburg, organized their third internal workshop that was held at the Fraunhofer Institute for Applied and Integrated Security (AISEC) Munich on 25-26th March 2025.

At the workshop, all partners reviewed the objectives and the ongoing project work. The discussion covered key topics such as project status updates, work package progress, and integration efforts, ensuring alignment with the project’s objectives. The workshop concluded with a clear roadmap for the next steps and a dedication to advancing the project.

GI-WebTalk #3: Sicherheitskonzepte und Geopolitik

Fri, 01 Nov 2024 14:00:00 +0100

Im dritten WebTalk unserer Reihe „Cybersicherheit und technologische Souveränität“ widmen wir uns zunächst den völkerrechtlichen Aspekten der Cybersicherheit und analysieren, wie internationale Normen und Verträge den Schutz kritischer Infrastrukturen und die staatliche Handlungsfähigkeit im Cyberraum sicherstellen. Der zweite Vortrag wird Ansätze zur Attribution von geheimen Nachrichten im Cyberraum vorstellen und einige die Herausforderungen darstellen, die mit der Identifikation von Verantwortlichen für Cyberangriffe einhergehen. Diskutieren Sie mit Expertinnen und Experten über die rechtlichen Rahmenbedingungen und ethischen Implikationen, die eine souveräne digitale Zukunft maßgeblich beeinflussen.

Ziel der ELSI-Webtalks (Ethics, Legal and Social Implications) ist der wissenschaftlich getriebene Diskurs zu diesen Fragen. Die Webtalks werden von der Gesellschaft für Informatik (GI) e.V. in Kooperation mit dem Forschungsprojekt SOVEREIGN durchgeführt, das seit 2023 mit Bundesmitteln durch die Cyberagentur gefördert wird.

By loading external content you agree to the privacy policy of YouTube.

Cyberagentur extends funding for phase 3

Thu, 05 Sep 2024 00:00:00 +0200

On August 20, we presented our results of the second phase of the Cyberagentur’s HSK call. We received a lot of positive feedback and got attention from other government agencies. We are happy to announce that we made it to the next phase and will receive 10 million in funding over the next three years to pave the way for securing critical infrastructures. We are thrilled, excited, and humbled to have made it through despite the intense competition in this call. We also want to acknowledge that we are deeply grateful for the support of all contributors who have been instrumental in our success.

Our project aims to increase the resilience and sovereignty of critical infrastructures through open hardware, confidential computing, and the consequent use of as well as contribution to open-source software. The goal is to bring a holistic approach to cyber security so that attacks can be detected and prevented early on.

GI-WebTalk #2: Strategien und Attribution im Cyberraum

Fri, 16 Aug 2024 13:50:00 +0200

Der zweite Webtalk der Reihe „Cybersicherheit und technologische Souveränität“ widmet sich der Analyse aktueller Angriffsstrategien sowie der Attribution von Cyberangriffen aus sicherheitspolitischer Perspektive. In den Fachvorträgen werden sowohl die technischen Möglichkeiten zur Detektion und Abwehr von Cyberbedrohungen als auch die Herausforderungen diskutiert, die mit der Identifizierung der Angreifer verbunden sind. Besondere Beachtung finden Fragen der technologischen Souveränität und deren Bedeutung für die Entwicklung effektiver Abwehrmechanismen.

By loading external content you agree to the privacy policy of YouTube.

Presentation of paper at ARES PCSCI Workshop

Mon, 05 Aug 2024 13:00:00 +0200

On July 30, 2024, at the International Workshop on Physical and Cyber Security in Interdependent Critical Infrastructures (PCSCI), which is an EU Projects Symposium Workshop held with the 19th ARES Conference 2024 in Vienna, Austria, our team member, Dr. Anum Talpur, presented the vision of our project SOVEREIGN.

The event gathered many researchers and experts from around the world who work in computer and information security. The PCSCI workshop, where we presented our project concept, was aimed at bringing security experts together to discuss potential solutions for tackling critical infrastructure protection. As part of our talk, we presented a novel and holistic concept of our SOVEREIGN platform that widely addresses the problem of securing critical infrastructures.

The workshop gathered a significant audience working on protecting critical infrastructures who expressed very positive feedback regarding our project, particularly highlighting the comprehensive concept and broad team of experts from industry, academia, and research institutes we have to accomplish our goals.

Georg Becker, Thomas Eisenbarth, Hannes Federrath, Mathias Fischer, Nils Loose, Simon Ott, Joana Pecholt, Stephan Marwedel, Dominik Meyer, Jan Stijohann, Anum Talpur, Matthias Vallentin. "SOVEREIGN - Towards a Holistic Approach to Critical Infrastructure Protection". International Workshop on Physical and Cyber Security in Interdependent Critical Infrastructures (PCSCI). Vienna, Austria. 2024.

PDF BibTeX

Cyberagentur visiting SOVEREIGN in Hamburg

Fri, 19 Jul 2024 11:00:00 +0200

On July 16, 2024, Prof. Dr. Christian Hummert, head of the Agentur für Innovation in der Cybersicherheit (Cyberagentur), with Dr. Gerald Walther and Dr. André Müller visited the Universität Hamburg to meet up with the SOVEREIGN project members to get a hold of the current state of the project. The members presented a coherent modular architecture to detect and defend against advanced cyber threats.

Figure 1: SOVEREIGN’s low-latency FPGA firewall allows malicious traffic to be filtered in legacy KRITIS infrastructure.

One discussed topic is the attribution of threat actors based on software binaries. Additionally, the project also presented specialized hardware that allows it to be integrated into the existing legacy infrastructure of KRITIS to dynamically assess the risk and take action. Another addressed topic is the enrichment and correlation of existing alerts from Intrusion Detection Systems (IDS) to aid the detection of multi-stage attacks. The project members also addressed potential attack vectors of the SOVEREIGN architecture by enrolling attestation to all software components, thus detecting tampering attempts. The project outcomes enable KRITIS organizations to better protect their IT and Operational Technology (OT) from attacks.

GI-WebTalk #1: Fehlinformation im Netz und Attribution von Cyberangriffen

Wed, 17 Jul 2024 16:00:00 +0200

Research results in the field of cyber security can usually be used for both civilian and military purposes, which obviously results in a dual-use problem. In addition, the increasing use of AI systems also raises ethical questions in the field of cyber security and in the security sector in general. The use of published research results and open source technologies for criminal and terrorist purposes cannot be ruled out either. For example, knowledge of state-of-the-art protection technologies could be used by adversaries to develop new strategies to circumvent them, and the protection options themselves could be further developed without making them available to the general public.

The aim of the ELSI Webtalks (Ethics, Legal and Social Implications) is a scientifically driven discourse on these issues. The webtalks are organised by the Gesellschaft für Informatik (GI) e.V. in cooperation with the SOVEREIGN research project, which has been funded by the Federal Cyber Agency since 2023.

The first GI webinar focussed on the topics of misinformation on the internet and attribution of cyber attacks. These are important topics that we are addressing as part of our SOVEREIGN research project. Many thanks to our great speakers Prof. Dr. Judith Simon and Marit Hansen for their contributions, to Nikolas Becker from the Gesellschaft für Informatik (GI) for moderating and to Dr. André Müller from the Agentur für Innovation in der Cybersicherheit GmbH (Cyberagentur) for the introductory words.

Figure 1: First webinar on 17 July 2024 on the topic of misinformation on the internet and attribution of cyber attacks.

By loading external content you agree to the privacy policy of YouTube.