The Four Seats
The Four Seats is a way of organizing an AI project so that everyone knows their specific job. It ensures that the people who pay for the AI, the people who run it, the people who build it, and the people who use it all stay in their own lanes, while a fifth 'governance' role acts as a referee to make sure the AI stays helpful and doesn't start making its own rules.
A framework for institutional AI governance that maps the lifecycle of an AI system across four distinct accountabilities: Board (Direct), Manager (Operate), Practitioner (Build), and User (Floor). It enforces a strict separation of concerns—specifically isolating the 'Vault' (the data/model) from the 'Brain' (the logic/governance)—to ensure that the technical implementation remains subordinate to the intended institutional purpose.
The capability routes that refract the single reality ('build the Vault, keep the Brain, never let the Vault judge') per accountability; plus a fifth Assure/Governance seat - the Brain made into a function that verifies the Vault never quietly started judging. (C4AIL, Whitepaper IV.) SEAT NAMES RUN IN TWO REGISTERS, same as the rungs. TECHNICAL: Floor=User, Build=Practitioner, Operate=Manager, Assure=Compliance/Governance, Direct=Leadership. CUSTOMER: Floor=AI User, Build=Builder, Operate=Manager, Assure=Governance Owner, Direct=Leader. Only Build differs for a STRUCTURAL reason - 'Practitioner' became the customer L3 LEVEL on 2026-07-22, so it cannot also name a route in that register; the others are phrasing for a customer ear. The Assure seat had NO customer name until 2026-07-22, which is why the assessment's lead-gen front door offered only four segments and the AI-Governance and SecureAI course ladders were unreachable through it. (canon: vocab-registry.yaml terms.seats.)