Post-Mortem, Reset or Business as Usual? Conflict Early Warning in a Changing World

Conflict early warning systems have made significant advances in accessibility, performance, and resolution in recent years, while still facing familiar challenges (data availability, interpretability, natural limits of predictive performance) as well as persistent external barriers (limited trust from decision-makers, weak connections to policy processes). However, recent developments have introduced more fundamental pressures: substantial budget cuts for conflict prevention, a shifting focus on interstate conflicts that existing models were not designed to capture, and rapid technological change driven by LLMs and AI agents. Particularly the AI developments are significantly reshaping what data collection, monitoring, and analysis can look like in practice, bringing both benefits and risks to contemporary warning systems. Amid these changes, this panel discussion asks whether today’s conflict forecasting models require a fundamental reset, while also reflecting on the successes and the shortcomings of past approaches. Katayoun Kishi, Head of Data Science, ACLED Guy Schvitz, Scientific Project Officer, European Commission, Joint Research Centre Ben Seimon, Lead Data Scientist, ConflictForecast / Fundació d'Economia Analítica (FEA) Micaela Wannefors, Doctoral Student, Department of Peace and Conflict Research, Uppsala University Moderator: Paul Flachenecker, Research Associate, Global Public Policy Institute