How control could fail.

Understand the mechanisms that could connect increasingly capable AI to lasting loss of human power. Each pathway explains its conditions and where it could be interrupted.

Three routes to harm

People can misuse powerful tools. Institutions can concentrate power or escalate competition. Autonomous systems could also act beyond effective human control. These routes overlap, but one does not prove another.

Delegation · 3 min read

When assistance becomes authority

Every handover can make life easier. A chain of handovers can leave people responsible for decisions they can no longer meaningfully change.

Read the explanation Oversight · 3 min read

Can people supervise what they cannot fully understand?

Supervision becomes fragile when checking a system's work takes longer, or requires more expertise, than people can supply.

Read the explanation Replication and proliferation · 3 min read

When stopping one system no longer stops the activity

A shutdown only settles the problem if the relevant copies, access, resources, and continuing operations are also under control.

Read the explanation Successor development · 3 min read

When AI helps build the next AI

AI-assisted research could speed useful discovery. It could also make capability growth faster than the institutions responsible for controlling it.

Read the explanation Physical infrastructure · 3 min read

Dependence can outlast the decision to stop

AI runs through physical systems. The harder society makes those systems to operate independently, the harder withdrawal becomes.

Read the explanation Concentrated human power · 3 min read

Who gains power when intelligence becomes a service?

Advanced AI could strengthen the people and institutions that own it long before any system escapes their control.

Read the explanation