An AI assistant should feel intuitive without feeling intrusive. When an algorithm predicts something so specific that it seems to know your private thoughts, it can trigger a cognitive uncanny response: a sense of being watched rather than helped. This “predictive creepiness” emerges when systems cross an invisible boundary between anticipating needs and overreaching into perceived mental privacy. Effective human–computer collaboration depends on understanding where those boundaries lie, and deliberately designing systems to stay on the right side of them. Measuring these limits across different interaction contexts and workflows is key to sustained trust and engagement.