If a non-human signal ever becomes reply-worthy, the authority to answer cannot sit inside one model stack.
The reason is not simply safety language. It is institutional legitimacy. Interpretation models can help parse meaning, risk, and uncertainty, but the moment the same stack also recommends reply structure and effectively authorizes transmission, the protocol begins to look self-referential. Too much judgment collapses into one technical pipeline.
A durable response system will need separation between translation, review, and reply authority. Different bodies may still rely on shared tooling, but they cannot all inherit their final confidence from the same inference layer without weakening the claim that the decision was actually governed.
Contact protocol is often imagined as a communications problem. It may turn out to be a separation-of-powers problem first.