Once we opened AI up broadly, a particular question started arriving from every direction. Am I allowed to use it for this?
At first I treated these as policy questions and answered them as policy questions. Yes for this, no for that, here’s the guidance page. The volume never dropped. And slowly I noticed something odd: many of the people asking already knew the answer. The guidance was clear, they’d read it, and they were asking anyway.
They weren’t asking for permission. They were asking for reassurance.
Underneath “is this allowed” sat quieter questions people rarely said out loud. If I use AI and it makes a mistake, who wears it? If I produce something twice as fast, will you doubt the work, or expect twice as much? Will using AI make me look capable or lazy? Is it safe to admit what I don’t know?
Those are not policy questions. They’re fear, dressed in compliance language, because compliance language is the safest costume fear can wear at work. Answering them with policy is like answering a child’s “are we lost?” with grid coordinates. Technically responsive, completely beside the point.
So we changed what we said, and more importantly, what we did. Leaders using AI openly and talking about their mistakes did more than any guidance page. We made it explicit that accountability for outputs sits where it always sat, with the person who owns the work, and that using AI well, including catching its errors, is a skill we’d celebrate rather than a shortcut we’d tolerate. The permission questions faded, not because people memorised the policy, but because the anxiety under the questions had been addressed.
There’s a diagnostic here I now offer other executives. Listen to the questions your people ask about AI. Policy questions have policy answers. But if the same “allowed?” questions keep coming after the policy is clear, your organisation doesn’t have a clarity problem. It has a safety problem, and only leaders can fix that one.