Alexander Müller

Director of Safe AI Netherlands (SAIN), the national AI safety field-building organization with chapters in Groningen, Utrecht, and Amsterdam. Currently co-founding the AI Safety Hub Amsterdam at The Stack. I'm generally interested in making sure AI goes well, for now focusing on preparing the Netherlands for the risks of transformative AI.

About me View CV Research Get in touch

What I'm Currently Working On

Scaling Safe AI Netherlands

Directing Safe AI Netherlands (SAIN), funded with a $2M two-year grant. Built a team of 4 full-time staff (including me), 10 part-timers, and 30 volunteers, and expanding from Groningen, Utrecht, and Amsterdam to Eindhoven, Delft, and The Hague.

AI Safety Hub Amsterdam

Co-founding and co-directing a dedicated AI safety hub at The Stack: a new place for researchers, professionals, and organizations working on AI safety in the Netherlands.

Research Interests

Multi-Agent Game Theory

How game-theoretic frameworks can help us think about multi-agent alignment and cooperation, i.e., how to defeat Moloch. See this paper.

LLM Agent Evaluations

Measuring how LLM agents behave when deployed in realistic settings, including whether they break the law. See this paper.

AI Alignment

How do we make sure AI systems actually do what we want? The problem is harder than it sounds. See e.g., this post.

Safe AI Netherlands (SAIN)

I direct Safe AI Netherlands (SAIN), the Dutch field-building organization focused on preparing the Netherlands for the risks of transformative AI. SAIN has chapters in Groningen, Utrecht, and Amsterdam, with Eindhoven, Delft, and The Hague coming soon. In 2026, we received a $2M two-year grant; we now run on a team of 4 full-time staff (including me), 10 part-timers, and 30 volunteers. What we do:

Education

• Educating hundreds of people through our AI Safety, Ethics & Society course
• Governance and Technical tracks
• Discussion groups on Technical AI Alignment and Governance

Research

• Project Hub with 6 experienced supervisors and ~15 researchers
• Papers at NeurIPS and ICLR workshops, and more
• Regular hackathons with podium placements

Growth & Collaborations

• Co-founding the AI Safety Hub Amsterdam at The Stack
• New chapters in Eindhoven, Delft, and The Hague

Events & Talks

• Turn.io, EAGx Amsterdam, aiGrunn
• Samenwerking Noord
• Discussion evenings and social events

Active on LinkedIn, Substack, Instagram, and our website.

Publications

View all →

Parametric Open Source Games

Todorov, Napel, Müller · arXiv 2026

EU-Agent-Bench: Measuring Illegal Behavior of LLM Agents Under EU Law

Lichkovski, Müller, Ibrahim, Mhundwa · arXiv 2025

Uncovering Internal Prediction Mechanisms of Transformer-Based Chemical Foundation Models

Müller, Cardenas-Cartagena, Pollice · ChemRxiv 2025

From Steering Vectors to Conceptors: Compositional Affine Activation Steering for LLMs

Abreu, Postmus, Müller, et al. · 2025

Collective Deliberation for Safer CBRN Decisions: A Multi-Agent LLM Debate Pipeline

Müller, Golicins, Lesnic · AISIG & Apart Research, 2025

Writing