AI Safety Fundamentals: SAIN Utrecht, Autumn 2026 (4th edition) AI Safety Fundamentals is a four-week introductory course run by SAIN Utrecht (Safe AI Netherlands). It ran from 23 September to 15 October 2026. Each week had a one-hour lecture, about one hour of readings, and a discussion group. Together they gave participants a clear, evidence-based view of the risks of advanced AI and how they are being addressed. The course began with a high-level introduction to AI safety. It covered the field’s key terms, how fast AI capabilities are growing, the main actors shaping the field, and the difference between technical and governance solutions. The second week looked at what can go wrong in practice. This included risks already visible in society today and frontier risks such as loss of control, cyber-offence, and misuse for chemical, biological, radiological, and nuclear (CBRN) weapons, each illustrated with real incidents. The third week explored why AI safety is hard to achieve, looking at the funding gap between capabilities and safety research, disagreement among experts, slow regulation, and international race dynamics. In the final week, participants chose between two tracks. The technical track introduced alignment, reinforcement learning from human feedback (RLHF), jailbreaking, evaluations, and interpretability. The governance track focused on the EU AI Act’s risk-based approach, accountability for AI systems, the challenges of general-purpose AI, and the international AI safety landscape. By the end of the course, participants could explain core AI safety concepts and recognise and classify different types of AI risk. They also understood the economic, political, and structural barriers to safe AI, and had a basic knowledge of both technical safety methods and the EU’s regulatory framework. Throughout the course, they practised critical thinking and open, respectful discussion of complex and uncertain topics.