Straight answer
Do we need an AI policy?
Yes, but a short one that people actually follow. Assume your staff are already using AI tools, with or without permission. A policy that says which tools are approved, what data must never go in, and who signs off new uses protects the company more than a long document nobody reads.
How can the board tell whether it is under control?
Signs it is under control
- Staff can name the approved tools and the kinds of data they must not put into them.
- New uses of AI that touch customers or decisions go through a named person before going live.
- The policy has been reviewed in the last year and changed at least once.
Signs it is not
- The policy is a ban on public tools, with no approved alternative.
- Nobody can say how many people use AI tools at work, or for what.
- The policy was written by one function and the business has never seen it.
What is the one question to ask?
How do we know whether people are following the AI policy?
A strong answer
Some evidence of real use, such as which approved tools are used and by how many, plus a route for staff to ask about new uses and a record of what was asked.
A weak answer
“It is on the intranet and everyone signed the annual acknowledgement.”
What should an AI policy actually cover?
A useful AI policy answers four questions in plain language. Which tools may staff use, and on which accounts? What information must never be put into them, such as customer data, personal data or unpublished results? Which uses need approval before they start, usually anything that affects customers, prices, hiring or other decisions about people? And who is accountable when something goes wrong? Most of this can fit on two pages.
Beyond that, the policy should say who keeps a list of AI uses in the business and how often the board sees it. Some industries and countries have specific rules on AI. Check which ones apply to you and who in the company tracks them, rather than relying on the policy to do that alone.
What might the board be getting wrong?
Boards sometimes ask for a policy as a way of closing the subject. Once it is approved, nobody asks again. A policy is only as good as the approved tools and the approval route behind it. If the board asks management to ban public tools without funding a safe alternative, people will work around the ban, and the board will have made the risk harder to see.
The board should also be clear about its own appetite. A policy written without that guidance tends to be either too cautious to use or too loose to matter.
What should the board ask for instead?
Ask for the policy, the list of approved tools, and the list of AI uses that have been through approval, all together. Ask how staff learn about the policy and where they go with a question. Ask for one example of a use that was refused or changed, and why. That shows the policy is working, not just written.