~/about

I’m a PhD student at the Institute for IT Security at the University of Lübeck, working where AI for Security meets Security of AI.

  • Security of AI: seed and prompt stealing, backdoors, steganography and the security of agent tool calls.
  • AI for Security: vulnerability prediction, security patch detection and neural decompilation.
  • Program Analysis: scaling symbolic execution with genetic programming and gradient-based analysis.

Please reach out to collaborate 😃

~/experience

  1. Research Intern

    IBM Research, Dublin
    Jun 2026 – Sep 2026
    Researched the security of AI-based coding agents.

  2. Researcher

    University of Lübeck
    Apr 2021 – Present
    Started as a research assistant developing mathematical models of neural networks. Since 2023 PhD student at the Institute for IT Security.

  3. Freelance Web Developer

    Self-employed
    Dec 2013 – Aug 2024
    Developed and customized TYPO3 and WordPress extensions, front end and back end (PHP, SQL, JavaScript). Projects include easy-sprachreisen.de and gaming-pc.net.

~/skills

Generative AI

  • Transformers
  • Diffusion Models
  • Mechanistic Interpretability
  • PyTorch
  • Unsloth

Agentic AI

  • LangChain
  • LangGraph
  • LangSmith

Optimization

  • Training (Language) Models
  • Adversarial Examples
  • Genetic Programming
  • Contrastive Learning

Program Analysis

  • Symbolic Execution
  • SMT Solvers
  • Java Bytecode Analysis
  • Static Security Scanners
  • Compiler Techniques

~/education

  1. PhD in Computer Science (IT Security)

    University of Lübeck
    May 2023 – Present
    Expected completion in April 2027.

  2. MSc in IT Security

    University of Lübeck
    Oct 2020 – Dec 2022
    Graduated with honors (1.2). Thesis: HyJaVu: Using Hybrid Program Traces to find Java Vulnerabilities.

  3. BSc in Computer Science

    University of Lübeck
    Oct 2017 – Oct 2020
    Focus on web and data science. Thesis: Comparison of different supervised machine learning models for anomaly detection in network data, in cooperation with PPI AG.

~/media & podcasts

~/publications

Felix Mächtle, Giulio Zizzo, Anisa Halimi, Thomas Eisenbarth, Mark Purcell (2026). AI4Good @ NeurIPS 2026 Marauder's Map: Mapping the reachable argument space of LLM-generated tool calls. Trustworthy AI for Good Workshop (AI4GOOD @ NeurIPS 2026).
Cite OpenReview
Felix Mächtle, Gautham Sriram, Manjil Neupane, Nils Loose, Thomas Eisenbarth (2026). ICICS 2026 Rethinking Neural Decompilation: Candidate Selection as the Hidden Bottleneck. 28th International Conference on Information and Communications Security (ICICS 2026), Fukui, Japan.
Cite DOI
Felix Mächtle, Jonas Sander, Sebastian Berndt, Ben Weimar, Nils Loose, Thomas Eisenbarth (2026). ARES 2026 Steganography Without Modification: Hidden Communication via LLM Seeds. Availability, Reliability and Security (ARES 2026).
Cite DOI arXiv
Nils Loose, Joseph Bienhüls, Kristoffer Hempel, Felix Mächtle, Thomas Eisenbarth (2026). ASE 2026 A Comprehensive Evaluation of Code Language Models for Security Patch Detection. Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE 2026).
Cite DOI
Nils Loose, Jonas Sander, Felix Mächtle, Thomas Eisenbarth (2026). NeurIPS 2026 FloatDoor: Platform-Triggered Backdoors in LLMs. Advances in Neural Information Processing Systems (NeurIPS 2026).
Cite arXiv Code Project
Gianluca Barmina, Federico Torrielli, Sven Harms, Jacob Nielsen, Felix Mächtle, Stine Lyngsø Beltoft, Peter Schneider-Kamp, Thomas Eisenbarth, Lukas Galke Poech, Anne Lauscher (2026). Preprint PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models.
Cite arXiv
Felix Mächtle, Nils Loose, Tim Schulz, Florian Sieck, Jan-Niclas Serr, Ralf Möller, Thomas Eisenbarth (2026). ASIACCS 2026 Trace Gadgets: Minimizing Code Context for Machine Learning-Based Vulnerability Prediction. 21st ACM ASIA Conference on Computer and Communications Security (ACM ASIACCS 2026).
Cite DOI arXiv
Felix Mächtle, Ashwath Shetty, Jonas Sander, Nils Loose, Sören Pirk, Thomas Eisenbarth (2026). ASIACCS 2026 Prompt Pirates Need a Map: Stealing Seeds helps Stealing Prompts. 21st ACM ASIA Conference on Computer and Communications Security (ACM ASIACCS 2026).
Cite DOI arXiv
Nils Loose, Florian Sieck, Felix Mächtle, Thomas Eisenbarth (2026). TACAS 2026 SWAT: Improvements to the symbolic executor (competition contribution). Tools and Algorithms for the Construction and Analysis of Systems – 32nd International Conference (TACAS 2026).
Cite DOI
Felix Mächtle, Jan-Niclas Serr, Nils Loose, Thomas Eisenbarth (2026). TACAS 2026 DASA: Fully Gradient-based Program Analysis (Competition Contribution). Tools and Algorithms for the Construction and Analysis of Systems – 32nd International Conference (TACAS 2026).
Cite DOI
Nils Loose, Nico Winkel, Kristoffer Hempel, Felix Mächtle, Julian Hans, Thomas Eisenbarth (2026). SBFT 2026 Coverage-Guided Multi-Agent Harness Generation for Java Library Fuzzing. Proceedings of the 19th ACM/IEEE International Workshop on Search-Based and Fuzz Testing (SBFT 2026).
Cite DOI
Felix Mächtle, Jan-Niclas Serr, Nils Loose, Thomas Eisenbarth (2026). DeepTest @ ICSE 2026 Beyond Accuracy: Characterizing Code Comprehension Capabilities in (Large) Language Models. Proceedings of the 7th IEEE/ACM International Workshop on Deep Learning for Testing and Testing for Deep Learning (DeepTest@ICSE 2026).
Cite DOI
Felix Mächtle, Jan-Niclas Serr, Nils Loose, Jonas Sander, Thomas Eisenbarth (2025). ACNS 2025 OCEAN: Open-World Contrastive Authorship Identification. Applied Cryptography and Network Security (ACNS 2025).
Cite DOI arXiv
Felix Mächtle, Nils Loose, Jan-Niclas Serr, Jonas Sander, Thomas Eisenbarth (2025). SBFT 2025 AutoStub: Genetic Programming-Based Stub Creation for Symbolic Execution. Proceedings of the 18th ACM/IEEE International Workshop on Search-Based and Fuzz Testing (SBFT 2025).
Cite DOI
Jens Alich, Thomas Eisenbarth, Hossein Hadipour, Gregor Leander, Felix Mächtle, Yevhen Perehuda, Shahram Rasoolzadeh, Jonas Sander, Cihangir Tezcan (2025). Preprint Attacks and Remedies for Randomness in AI: Cryptanalysis of PHILOX and THREEFRY.
Cite ePrint
Nils Loose, Felix Mächtle, Florian Sieck, Thomas Eisenbarth (2024). TACAS 2024 SWAT: Modular Dynamic Symbolic Execution for Java Applications using Dynamic Instrumentation (Competition Contribution). Tools and Algorithms for the Construction and Analysis of Systems – 30th International Conference, TACAS 2024, Held as Part of the European Joint Conferences on Theory and Practice of Software, ETAPS 2024, Luxembourg City, Luxembourg, April 6-11, 2024, Proceedings, Part III.
Cite DOI
Nils Loose, Felix Mächtle, Claudius Pott, Volodymyr Bezsmertnyi, Thomas Eisenbarth (2023). DIMVA 2023 Madvex: Instrumentation-Based Adversarial Attacks on Machine Learning Malware Detection. Detection of Intrusions and Malware, and Vulnerability Assessment (DIMVA) – 20th International Conference.
Cite DOI arXiv