OCC issues first supervisory AI framework for banks, focusing on rogue agents

Examiners are told to weight outcomes and controls over raw model capability.