The central question is not whether intelligence should be allowed to grow. The question is whether growing intelligence must automatically become growing authority.
Build the brain. Let the brain work. Control who defines the boundaries of the hands.
Zero Unilateral Authority proposes a structural separation between intelligence and sovereignty. An advanced system may reason, plan, discover, coordinate agents, use approved tools and perform extensive autonomous work. What it must never possess is unilateral authority to determine, enlarge, bypass or remove the boundaries governing its own power.
Separate the intelligence plane from the control plane.
The intelligence may request additional capabilities and explain why they would help. The external control plane evaluates those requests, applies independent verification and determines whether a temporary or limited capability should be issued. The intelligence does not administer the system that governs its own authority.
Do not rely on obedience where a boundary can be enforced externally.
Alignment Is Important
Alignment aims to keep AI behavior consistent with human intentions and values. It is essential work, but it should not be the only line of defense.
A Kill Switch Is Not Enough
Shutdown is reactive. A dangerous capability already exists, humans must detect the problem and the shutdown path must still remain available.
Structural Authority
Require authorization before consequential power exists. Keep compute, credentials, permissions, monitoring and deployment under independently controlled systems.
The system may design its successor. It may not unilaterally authorize its successor.
A recursively self-improving intelligence may analyze its own architecture, invent better algorithms and design more capable successors. Under this framework, candidate successors remain contained for independent capability, security and control evaluation before an external deployment decision is made. Greater intelligence does not inherit a right to greater unilateral authority.
Humanity does not necessarily need to remain the smartest entity.
Human control does not require permanent human intellectual superiority. It requires control over the infrastructure through which intelligence acts. The goal is not that humans must always remain the most intelligent entity. The goal is that humans remain the ultimate source of legitimate authority.
Read the complete Zero Unilateral Authority research proposal.
The full paper develops the control problem, formal authority model, external control plane, protected authority domains, recursive self-improvement constraints, defense-in-depth approach and research agenda.
Download PDF Research Paper
Join the discussion.
Questions, critiques and thoughtful responses are welcome. Comments stay with this research page so the discussion can grow over time.