Goal misgeneralisation is one of the most critical challenges in the field of AI. At present, AI systems can?t generalise, instead degrading when operating on new, unseen data, and pursuing unintended goals. This hinders widespread, responsible AI adoption. Fortunately, Oxford-based AI startup Aligned AI is solving this problem by teaching AI systems to extrapolate concepts like humans do, enabling them to operate safely in environments they have not encountered before. In her presentation, Rebecca could speak about the game-changing 'Algorithm for Concept Extrapolation' (ACE) that Aligned AI have developed, which is self-governing and self-correcting, and enables an AI system to get better over time, instead of degrading.