Who’s actually accountable if your AI goes wrong?

Most board conversations about AI cover similar ground. What it could do for the business. Which competitor’s already using it. How fast the market’s moving. All genuinely useful, worth having.

There’s often a smaller question sitting underneath them, one that’s easy to skip because it doesn’t feel urgent yet. If this goes wrong, who’s actually accountable?

Not IT. Not “the team.” An actual person whose job it is to have thought about it before it happens.

Approving the budget isn’t quite the same as owning the risk

Most boards are pretty comfortable signing off on AI spend by now. A new tool here, a Copilot rollout there. That part feels familiar, because it looks like any other technology decision a board’s made before.

The accountability question underneath it is newer ground for most of us. A finance system has an owner. A data protection policy has an owner. AI, in a lot of businesses, doesn’t quite yet. It tends to sit somewhere between IT, operations, and whoever asked for the tool in the first place, and ownership can drift without anyone quite meaning it to.

That gap tends not to matter until something goes wrong. Client information ends up somewhere it shouldn’t. An AI-generated answer gets used without being checked. Usually nobody set out for that to happen, it’s just what can occur when nobody’s specifically watching for it.

A few honest questions worth asking

A board doesn’t need to understand the technology to do this well. A short list of honest questions and a willingness to sit with the real answers goes a long way.

Which tools are people actually using day to day, and were any of them approved. What happens to client data once it goes into an AI system. Who checks AI-generated work before it reaches a client or a filing. Who’s the named person accountable if something goes wrong.

If the honest answer to any of those is “we’re not sure yet,” that’s a perfectly normal place for a lot of businesses to be right now. It’s just worth knowing, so it can be fixed before it becomes something more than a question.

What an unclear answer can end up costing

When this goes wrong for a business, it’s rarely AI itself that’s really the problem. More often it’s that nobody had quite decided anything, so nobody was in a position to catch it early.

Fabricated information ending up somewhere it shouldn’t. Confidential material sitting in a tool nobody’s had a chance to vet, waiting to resurface later in ways nobody planned for. None of this needs AI to be malicious, or even particularly advanced. It just needs a gap nobody’s had time to close yet.

Building it in before you need it

The fix isn’t complicated, and it doesn’t need to slow anything down. A named owner. A short list of approved tools. A plain answer on client data, one you’d actually be comfortable saying out loud to a client. Once those exist, the board’s job gets a lot easier, there’s something real to check rather than a general sense that AI is “being looked at.”

Confidence with AI isn’t really about moving carefully. It’s about knowing who’s accountable if things don’t go to plan, and having had that conversation before you needed the answer.