I keep returning to a structural question:
In Amodei's framework, safety is achieved through:
• Internal constitutions (Constitutional AI)
• Corporate governance (RSP, ASLs)
• Government regulation (Advanced AI Framework)
Three institutional layers. All centralized. All dependent on competent, uncorrupted actors.
What happens when one layer fails? When a government is captured? When a corporation faces quarterly pressure?