Research Group
Machine Learning
and Security
and Security
View from our building over Berlin.
Our research group conducts fundamental research at the intersection of computer security and machine learning. On the one end, we are interested in developing intelligent systems that can learn to protect computers from attacks and identify security problems automatically. On the other end, we explore the security and privacy of machine learning by developing novel attacks and defenses.
We are part of the Berlin Institute for the Foundations of Learning and Data (BIFOLD) at Technische Universität Berlin. Previously, we have been working at Technische Universität Braunschweig and the University of Göttingen.
November 13, 2026 — We are hiring! We have a new PhD position in our project AMIRA on the security of mechanistic interpretability. The application deadline is November 26, 2026.
August 10, 2026 — We are attending the USENIX Security Symposium in Baltimore, 🇺🇸. Josiane is presenting our paper on personalized phishing using large language models.
June 3, 2026 — We are attending AsiaCCS in Bangalore, 🇮🇳. Felix is presenting our paper on shape-shifting malicious code using large language models.
See all news and updates of the research group.
When a Zero-Shooter Cheats: Improving Age Estimation via Activation Steering.
Advances in Neural Information Processing Systems 40 (NeurIPS), 2026.
Fingerprinting Inference Systems of Large Language Models.
Advances in Neural Information Processing Systems 40 (NeurIPS), 2026.
On the Same Track: Privacy Leaks in Electric Rail Transport via Magnetic Fields.
33rd ACM Conference on Computer and Communications Security (CCS), 2026.
Secrets Everywhere: Auditing Memorization in Mobility Prediction Models.
33rd ACM Conference on Computer and Communications Security (CCS), 2026.
See all publications of the research group.
AMIRA — Adversarial Mechanistic Interpretability
The project explores the security of mechanistic interpretability. It investigates how adversaries can manipulate models and inputs to induce fabricated circuits and false explanations, and derives defenses for detecting and hardening against such attacks. The project is joint work with KIT and Imperial College London.
DISTEL — Differential Security Testing of LLMs
This project introduces differential security testing for LLMs. It aims to compare LLM behavior across model variants, software stacks, and hardware platforms. The goal is to assess whether differences can impact the security and trustworthiness of LLMs. The project is part of the excellence cluster CASA.
MALFOY — Machine Learning for Offensive Computer Security
The ERC Consolidator Grant MALFOY explores the application of machine learning in offensive computer security. It is an effort to understand how learning algorithms can be used by attackers and how this threat can be effectively mitigated. The project develops novel attacks and corresponding defenses.
See all projects of the research group.
BIFOLD & TU Berlin
Machine Learning and Security (FR7-4)
Franklinstraße 28-29
10587 Berlin, Germany
Office: office@mlsec.tu-berlin.de
Responsibility under the German Press Law §55 Sect. 2 RStV:
Prof. Dr. Konrad Rieck