A shared language for democratic AI

The Democracy Levels Framework was developed to help AI organizations, regulators, and stakeholders integrate meaningful democratic processes into AI governance and alignment. It provides concrete language and practical tools for determining when and how to devolve decision-making power.

The framework defines six levels of democratic decision-making, from L0 (fully unilateral) to L5 (full democratic governance). Each level describes who has decision-making power and how decisions are made.

L0: Unilateral decision-making L1: Democratic processes inform decisions L2: Fully specified democratic outputs (vetoable) L3: Binding democratic decisions L4: Automatically triggered binding processes L5: Democratic metagovernance with checks and balances