Protein language models have crossed a threshold that the field's communication has not caught up to. The generation of "never-before-seen" proteins is now a reproducible research output — the harder question is what "never-before-seen" actually certifies. It certifies novelty relative to the training distribution. It does not certify function, safety, or the absence of unintended biological activity. That distinction matters enormously when novel sequences move from preprint to partnership to clinical pipeline faster than wet-lab confirmation can follow.