Europe's first ELLIS Institute, based in Tübingen, Germany, conducting pioneering fundamental AI research with dedicated groups in AI safety, alignment, and robust machine learning.
Europe's first ELLIS Institute, based in Tübingen, Germany, conducting pioneering fundamental AI research with dedicated groups in AI safety, alignment, and robust machine learning.
People
Updated 05/18/26Scientific Director
Managing Director
Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- $125,000,000
Org Details
Updated 05/18/26The ELLIS Institute Tübingen (ELLIS Institute Tübingen gGmbH) is Europe's first and the world's first institute within the ELLIS (European Laboratory for Learning and Intelligent Systems) network. Its founding was announced in January 2022 when the Hector Foundation and the state of Baden-Württemberg signed a funding agreement, and it officially launched operations in July 2023. The grand opening ceremony took place on June 20-21, 2024, during Cyber Valley Days, attended by dignitaries including Science Minister Petra Olschowski, Turing Award winner Yann LeCun, and the institute's Scientific Director and ELLIS President Bernhard Schölkopf. The institute is located at Maria-von-Linden-Straße 2, 72076 Tübingen, Germany, embedded within the broader Tübingen AI ecosystem that includes the Max Planck Institute for Intelligent Systems and the Tübingen AI Center. Almost all principal investigators hold co-affiliations at one or both of these partner institutions. As of early 2026, the institute hosts approximately 12 research groups led by Hector Endowed ELLIS Fellows. Research groups span AI safety and alignment (Maksym Andriushchenko), cooperative machine intelligence for people-aligned safe systems (Sahar Abdelnabi), safety- and efficiency-aligned learning (Jonas Geiping), robust machine learning (Wieland Brendel), AI mechanisms (Rediet Abebe), algorithms and society (Celestine Mendler-Dünner), AutoML (Frank Hutter), deep models and optimization (Antonio Orvieto), computational applied mathematics and AI (T. Konstantin Rusch), science and probabilistic intelligence (Maximilian Dax), wild/efficient/innovative AI (Shiwei Liu), and empirical inference (Bernhard Schölkopf). The institute plans to recruit up to 15 principal investigators total. Total personnel including PhD students, postdocs, and administrative staff numbers approximately 130. Notable recent achievements include: Maksym Andriushchenko receiving a $1,000,000 grant from Coefficient Giving for AI safety research; a two-year grant from the UK's Department for Science, Innovation and Technology (DSIT) awarded jointly to Sahar Abdelnabi, Maksym Andriushchenko, and Jonas Geiping to develop open-source benchmarks and intervention techniques to detect AI systems that game their own evaluations; and Andriushchenko receiving the ELLIS Best PhD Award 2025. The institute is part of a pan-European ELLIS network encompassing over 40 units across 14 countries.
Theory of Change
Updated 05/18/26The ELLIS Institute Tübingen believes that maintaining Europe's sovereignty and leadership in foundational AI research, conducted within a framework of European values (human-centered, beneficial, and safe AI), is critical to ensuring AI development goes well. By attracting world-class machine learning talent and providing excellent research conditions, the institute generates fundamental scientific advances that improve our understanding of how modern AI systems work and how to make them safer. Multiple research groups directly address AI safety risks: the AI Safety and Alignment group develops technical solutions to reduce risks from general-purpose AI models, focusing on alignment of autonomous LLM agents and rigorous AI evaluations to assess frontier model capabilities and risks; the COMPASS group works on safe, aligned, interpretable, and steerable AI agents; and the safety-aligned learning group studies optimization for secure systems. The causal chain is: excellent fundamental research and talent development → better scientific understanding of AI systems → technical tools for alignment, evaluation, and detecting deceptive AI behavior → reduced risk from advanced AI systems as these findings inform how AI labs and policymakers build and govern frontier models.
Grants Received
Updated 05/18/26Projects
Updated 05/18/26
Grant-funded collaboration in which Coefficient Giving provides a $1,000,000 grant to support PI Maksym Andriushchenko’s AI safety research on safe and aligned autonomous LLM agents.
Education-focused project to transform the “MatheBattle” platform into an open-source, adaptive mathematics learning environment that integrates self-regulation prompts and AI-driven tutoring to strengthen autonomous learning.
Collaboration with the UK AI Security Institute’s Alignment Project to study AI models’ test awareness and develop methods to detect and reduce evaluation gaming and security risks in advanced AI systems.
Discussion
No comments yet. Be the first to share your thoughts.