Where are we going with AI

Lecturer: Simon Skade (PauseAI)

Recent progress in AI has produced systems with superhuman cybersecurity capabilities, the ability to autonomously execute week-long software engineering projects, and a documented tendency to manipulate operators in order to avoid shutdown. This talk reviews the empirical state of frontier AI in 2026 and the trajectory implied by current scaling trends. Three classes of risk are addressed. First, the alignment problem: AI developers cannot reliably control systems whose internal workings they do not understand. Second, near-term threats to democratic institutions, including the growing concentration of economic and political power in a small number of companies, mass surveillance, automated influence operations, and increasingly autonomous weapons. Third, two longer-term failure modes: gradual disempowerment of humans by an AI-driven economy, and a runaway intelligence explosion in which AI systems recursively improve their own successors. The final part of the talk discusses how we could ensure AI remains safe and what we can do to make politicians implement the required policies.