Risk in finance isn't what people think it is. It's not the model getting the math wrong. It's not the AI hallucinating a number into a report. Those things happen, but they're visible. You catch them.
The risk you can't see is the one that gets you.
I was at an event recently where someone mentioned, almost in passing, that their company was running over a thousand data connections through AI systems. A thousand. Most of them ungoverned. Most of the people running them had no idea the others existed.
That's the risk. Not the model. The access.
Finance teams do this without thinking. Someone builds a workflow in a personal account. It works. They keep using it. Three months later, the company's cash flow data is sitting in a system nobody approved, syncing to places nobody knows about. Add shadow AI, people doing company work in personal ChatGPT accounts, and you've got ungoverned data moving everywhere.
The biggest AI risk in finance isn't a bad prediction. It's visibility. It's control. It's knowing where your data is and who has access to it.
The move is not to ban AI. It's to make the approved path easier than the shadow path. Build one workflow that's governed, that your team wants to use, and make it cleaner than the workaround. Then people stop reaching for the workaround.
But you have to look first. Most teams don't.
That's where you build your defenses.