Top AI executives are openly split on whether development should slow for stronger safety checks. The rift underscores a high-stakes struggle over who should set guardrails for increasingly powerful systems.