Aligning the model was never going to govern it

Most AI roadmaps rest on a quiet bet: that the labs will eventually ship a model safe and aligned enough to simply trust in production. Better training, better guardrails, one more version, and the thing behaves. Here is the flaw in that…