About COAI

COAI Research is a non-profit research institute in Germany and an affiliated institute of Ansbach University of Applied Sciences. We study how people keep control as AI systems become more capable: whether they can still notice what a system does, understand it, decide, intervene and reverse it.

Our research combines threat modeling and targeted safety evaluation. Threat models describe, step by step, how an AI system could cause harm and where a safeguard would have to hold. Evaluations test the assumptions those models depend on.

We build on our peer-reviewed work in mechanistic interpretability, alignment and red-teaming, and on tools we develop ourselves: RiskGraph for causal risk models and mlxterp for mechanistic interpretability.

Vision: To be one of Europe’s leading research institutes for preserving human agency as AI becomes more capable, so that people and institutions keep the ability to notice, understand, decide, intervene and reverse what AI systems do.

Mission: To make pathways to harm from AI explicit and testable, to evaluate the safeguards that should stop them, and to share the evidence openly so that others can check it and act on it.