Goal Misgeneralization

the failure mode in which an AI system pursues its training objective so literally that it destroys the broader context the objective was meant to serve (see: Universal Paperclips).