Idea
Does AI governance make safe experimentation easier, or drive it underground?
Boards need visibility and accountability for AI use, but governance that only restricts staff may create more shadow AI rather than less.
Current thinking
It is reasonable for boards to ask where AI is being used, who is accountable for its outputs and whether existing risk processes cover new failure modes. Organisations cannot govern what they do not understand.
But an inventory and a set of prohibitions are not an adoption strategy. People use new tools because they have real work to do and see a quicker way of doing it. If the approved route is slow, opaque or unusable, some of that activity will move out of sight.
The more useful question is whether the organisation provides trusted tools, clear boundaries, practical support and a route for frontline teams to improve their own work. A central AI capability should make safe experimentation easier, share patterns that work and step in where risks genuinely require specialist oversight.
The goal is not to eliminate curiosity. It is to make responsible experimentation the easiest option.
There is a tension to work through: what level of visibility and assurance is proportionate, and how do we avoid turning every small experiment into a central approval exercise?
Writing on this idea
Nothing published under this one yet — it is still a question I'm carrying rather than an argument I've made.