AI Governance

This framework defines the minimum set of governance, engineering and evidentiary requirements that an organisation shall satisfy in order to deploy, operate, study or rely on artificial intelligence systems in a manner that is accountable, auditable and defensible. The central premise is unchanged from the framework it supersedes. Artificial intelligence systems are not autonomous decision makers. They are a supervised, constrained and auditable processing layer that accelerates human cognitive work without substituting human accountability. To that premise this edition adds a second one, which governs everything in the evidence domain. A statement produced by, or about, an artificial intelligence system is not established because the system produced it, because a metric improved, or because a demonstration was persuasive. It is established when a documented procedure exists under which the statement could have failed and did not.