On news platforms, a user clicks on an article based on her subjective anticipation of its content, inferred from the title using her background knowledge. We introduce SAND (Subjective Anticipation-augmented News Recommendation via Dual-pathway Distillation), the first framework to explicitly incorporate this user-specific anticipation that drives click behavior, formalized as the anticipated body, into news modeling. SAND (i) generates personalized anticipated bodies via an agent-based collaborative system, and (ii) learns preferences from anticipated bodies using a candidate-aware attention network with dual-pathway knowledge distillation. Experiments on real-world news datasets (i.e., MIND and Adressa) demonstrate that SAND significantly outperforms eleven state-of-the-art baselines.