Two years ago, board discussions concerned capability. The current agenda is accountability.
Directors increasingly ask three questions. Which decisions are delegated to a system, and under what authority? What evidence exists that a given output can be reconstructed and explained? And where does liability sit when an automated decision is wrong in a regulated process?
These questions favour deterministic and traceable designs over impressive but opaque ones, and they favour vendors willing to be specific about logging, model provenance and exit paths.
The organisations answering them well tend to treat AI governance as an extension of existing operational risk practice, rather than a separate technology initiative.