· Blog
Is the AI governance firms report the one that actually governs?
Annual reports describe AI oversight in remarkably similar terms. I am studying what happens between that page and the place where AI is actually used.
Read enough annual reports and you start to notice that AI governance has a house style. Committees, principles, responsible-AI frameworks, board oversight. The language travels from company to company, and from continent to continent.
Institutional theory has a name for this kind of convergence, and a long tradition of explaining the gap between what organisations display and what they do. But AI raises a new possibility. What if the gap is not about ceremony at all, but about architecture? What if firms cannot govern certain risks simply because those risks originate in models that someone else builds?
How I study it
I follow corporate disclosure over a decade, across firms in Asia, Europe, North America and Africa, and I compare what those documents say with what executives and advisers describe when they talk about how AI is really used. The question is not whether firms are honest. It is where governance can reach, and where it cannot.