AISIG Research Hub

Connecting supervisors, collaborators, and students to advance AI safety research

At AISIG, we care about high-quality research.

Our Research Hub offers a platform where supervisors, collaborators, and students can connect to advance AI safety research on a global scale.

There are two ways in which you can participate:

(1) Supervised Research Matching: Apply to get supervised by or become one of our experienced researchers (more info below).

(2) Open Collaboration Opportunities: Apply to collaborate on research, either alone or in a team (more info below).

Supervised Research Matching

Students and early-career researchers can apply for guided research projects. AISIG provides logistical and financial support (e.g. compute). These projects aim to evolve into research worthy of publishing (some of the projects at the bottom of this page are examples).

Apply here

Meet Our Supervisors

Experienced researchers (PhD+) with an interest in AI safety research can join AISIG’s Supervisor Team. We welcome supervisors from any discipline.

Email us at info@aisig.org.

Open Collaboration opportunities

Not all research requires formal supervision. AISIG connects researchers directly and provides support (e.g., funding for compute). We streamline collaboration so you can focus on advancing AI safety. If you have an interesting (research) idea, then kindly submit your idea with the below form. If you want to join a project, please reach out to the contact person of the particular project you’re interested in.

Submit ideas

We also welcome you to join one of the research efforts below:

Selected research from our members

  • Steering Large Language Models using Conceptors (NeurIPS, 2024)

    Joris Postmus and Steven Abreu [link]

  • Self-Ablating Transformers: More Interpretability, Less Sparsity (ICLR, 2025)

    Jeremias Ferrao [link]

  • Contextual Sparsity as a Tool for Mechanistic Understanding of Retrieval in Hybrid Foundation Models (ICLR, 2025)

    Davide Zani, Felix Michalak, and Steven Abreu [link]

  • Playing with Perception: Fooling Traffic Sign Classifiers via Copy-Paste Manipulation

    Davide Zani and Alexandru Dimofte [link]

  • Sandbagging LLMs using Activation Steering

    Jeremias Ferrao and Davide Zani [link]

  • Cybersecurity Persistance Benchmark

    Davide Zani, Felix Michalak and Jeremias Ferrao [link]

  • AI Misinformation and Threats to Democratic Rights

    Davide Zani, Mariam Ibrahim, Tiwai Mhundwa, Felix Michalak, and Andrei Avram [link]

  • AutoSteer: Weight-Preserving Reinforcement Learning for Interpretable Model Control (1st place Apart Research Hackathon)

    Jeremias Ferrao [link]

  • Local Learning Coefficients Predict Developmental Milestones During Group Relative Policy Optimization (3rd place Apart Research Hackathon)

    Jeremias Ferrao and Ilija Lichkovski [link]

  • Collective Deliberation for Safer CBRN Decisions: A Multi-Agent LLM Debate Pipeline (4th place Apart Research Hackathon)

    Alexander Müller, Arsenijs Golicins, and Galina Lesnic [link]

  • The Anatomy of Alignment: Decomposing Preference Optimization by Steering Sparse Features (NeurIPS spotlight paper, 2025)

    Jeremias Ferrao, Matthijs van der Lende, Ilija Lichkovski [link]