AI is already in your organisation. Govern it like it is.

In almost every organisation I speak to, the official position on AI is still being drafted. Meanwhile, staff are pasting meeting notes into chat assistants, suppliers are switching on AI features inside platforms you already pay for, and developers are wiring models into internal tools. The policy is catching up with something that has already happened.

That is not a criticism of anyone. It is how useful technology always arrives: through people trying to do their jobs better. But it means the real question is no longer whether to adopt AI. It is whether you can see where it is already being used, and whether anyone owns the risks.

The risk is mostly familiar

There is a lot of noise about exotic AI threats. Most of the exposure I actually see is old risk in new clothes: sensitive data going to a third party without a contract, decisions being made on outputs nobody checked, and access granted to a tool far wider than the task needs.

• Data leakage — personal, commercial or classified information shared with services outside your control.
• Supplier risk — AI features turned on by default in existing platforms, often without a fresh assessment.
• Over-trusted output — summaries, code and advice used without a human who is accountable for checking them.
• Excessive access — AI assistants connected to email, files and systems with the permissions of whoever set them up.
• Unclear accountability — when something goes wrong, nobody can say who approved the use case.

You do not need a new risk framework for AI. You need to apply the one you have to the places AI has already reached.

Why banning it does not work

The instinctive response is to block everything until the policy is ready. In my experience that rarely reduces risk. It pushes use onto personal devices and personal accounts, where you have no visibility at all. People do not stop needing to work faster because a tool was blocked.

A better approach is to offer a safe, approved route that is genuinely easier than the unsafe one, and to be clear about the small number of things that are never acceptable. Guardrails people can live with beat prohibitions they quietly ignore.

Five questions for leadership this quarter

• Where is AI already being used across the organisation — including inside the software we already buy?
• What information must never go into an AI tool, and do our people know that in plain language?
• Which AI use cases affect customers, citizens or staff decisions, and who is accountable for each one?
• Have our critical suppliers told us how they use AI with our data, and have we asked?
• If an AI-assisted decision turned out to be wrong, could we explain how it was made?

None of these need a specialist to answer. They need someone with authority to ask them, and the patience to get honest answers rather than reassuring ones.

Keep people at the centre

The organisations handling this well are not the ones with the longest AI policy. They are the ones that treat AI as something people use, not something that happens to them. They train staff on judgement, not just rules. They keep a human accountable for every output that matters. And they make it easy to ask, "Is this okay?" before something goes wrong rather than after.

AI will change how we work, and much of that change will be good. The job of security and risk professionals is not to stand in its way. It is to make sure that when it arrives, it arrives with an owner, a boundary and a way back if things go wrong.