The week a written rule goes live, the questions do not arrive at the legal inbox first. They arrive in a conversation a manager already has. A person is holding a file and asking whether it can go into the tool, or whether a draft that used the tool can go to a customer. If that manager has only seen the policy as a slide, the answer will be a guess, and the guess will travel further than the document.
Microsoft's guidance on how to create an AI policy employees can follow, published on 12 August 2026, says staff should be able to find what they can do, what they cannot do, why the rules exist, where the approved tools are, and who to contact when a question is not covered. The same note puts the team manager on the path for an AI-generated error, next to an AI governance lead. The person closest to the work will be asked first, including after something wrong has already gone out.
The questions worth rehearsing
Turn those categories into questions your teams already ask. Can I paste this customer brief. Can I use a tool that is not on the list. Do I have to say this draft was helped. What do I do if a number was wrong and the email has gone. The manager has to answer these on a Tuesday without hunting for a PDF.
Microsoft's restricted list covers confidential customer information, regulated data without authorisation, a final decision with no review, and an official communication with no approval. Ordinary uses on the same page include a first draft, a summary, help with analysis and help with code. Your lists will differ. Rehearse yours. A manager who has only practised a generic list will hesitate, and the person asking will hear the hesitation as a yes.
Have each manager say the answer out loud, with the policy on the table and a colleague playing the person who asked. The answer has to be short enough for a corridor. Allowed, not allowed, or allowed with a check, plus the name of the person who takes whatever the policy does not cover. If the manager cannot say that name, the policy is not ready, and another all-staff note will not fix it. A new tool goes to the technology or security team. A data question goes to legal or security. A wrong output goes to the manager or the governance lead. A committee that meets next month is not an answer for a file that has to leave today.
What to look at the following week
Collect five real questions in the first week, from the managers, not from a survey the project team designed. Look at whether the answers matched the policy, whether the escalation name was used, and whether any question had no answer at all. A question with no answer is a hole in the policy. Patch the policy and tell the managers the patch in the same week. Do not collect the questions and file them.
Staff need to know the line. Managers need to apply it when the case is messy, and to know when they are no longer the right person to answer. Book the manager hour before the all-staff note, not after the first confused week. A manager who meets the question for the first time in public will improvise, and the team will repeat that improvisation as if it were the rule.
Collective Campus is building high-performance culture and teams for the AI world. The AI governance workshop is where leaders and the people who will be asked the questions sit with the policy and practise the answers. What the staff session should cover is in acceptable use training for staff. A policy nobody can find is covered in the findable policy problem. Files pasted before a rule exists are covered in the brief that went in early.
What to bring
Bring the policy, the approved tool list and five questions you have already heard. If you have not heard any, ask three managers this week what their teams are already doing. Those answers are the curriculum.
