Our paper "Pinpointing Attention-Causal Communication in Language Models" has been accepted at NeurIPS 25.