Datum: 13 maj 2026
Tid: 10:00-11:45
Venue: Institute for Futures Studies, Holländargatan 13 in Stockholm or online
Research seminar with Carina Prunkl, Research Scientist at Inria’s REGALIA project and a Senior Research Fellow at the University of Oxford’s Institute for Ethics in AI. She is also Lead Writer of the 2026 International AI Safety Report and Co-Investigator of the UKRI funded project CHAILD. Her main research focus is on agency, autonomy and human oversight, though she is also interested in the governance of AI more broadly.
Abstract
This presentation discusses key findings from the 2026 International AI Safety Report, an international scientific assessment of the capabilities, risks, and risk management challenges associated with general-purpose AI systems. It focuses in particular on one cross-cutting theme in the report: the growing gap between what current evaluations measure and what decision-makers need to know. As AI systems advance rapidly, benchmarks and model evaluations are increasingly used to support claims about capability, safety, and readiness for deployment. Yet these evaluations often provide only limited evidence about how systems will behave in real-world settings. The talk situates this evaluation gap within the report’s broader findings on capability advances, emerging risks, and the limits of existing safety measures, and reflects on what this means for governance under conditions of incomplete and uneven evidence.
If you wish to receive our newsletters, and invitations to our seminars, subscribe here.
