Policy
159d agoApproval‑Directed Agents: A Policy Blueprint
Approval‑directed agents promise a new governance model for advanced AI, ensuring systems act only when aligned with human intent. The framework, pioneered by Paul Christiano through Iterated Amplification and Iterated Distillation and Amplification, could standardize safety protocols worldwide.
The Signal
By making AI behavior transparent and accountable, it offers a scalable path to mitigate existential risk across borders.