A coordinator has a review at five and a page that is still a pile of notes. The approved list of tools, if anyone wrote one, sits on the intranet. The chat window on the laptop is already signed in, and the notes go there because the draft has to exist before the meeting does.
I see the same hour in teams that already have a written rule and in teams that are still arguing about one. The work does not pause for the document. Company material and client material live in the same folder as the task. A person who is trying to finish will use the tool that answers, and they will use the material they already have open.
The habit is ahead of the rule
Microsoft's 2024 Work Trend Index, produced with LinkedIn, found that 75 percent of knowledge workers use AI at work, and that 78 percent of AI users bring their own AI tools. The report says this approach puts company data at risk, in a year when leaders named cybersecurity and data privacy as their leading concern. Only 39 percent of people who use AI at work had received training from their company.
The same research says 52 percent of people who use AI at work are reluctant to admit they use it on their most important tasks. A hidden habit and a late rule are a poor pair. The file still goes in. The person simply stops telling you, which means the risk never makes the agenda.
A document that arrives after that habit has a hard job. It can name principles. It can point at a committee. It will not, on its own, change the five o'clock decision, because the person with the deadline was not in the room that wrote it. AI governance training is how you put that decision in front of the people who touch the file.
Company data and client data
Microsoft names the risk as company data. On the desks I sit beside, the file is often a client's. A brief, a contract draft, a spreadsheet of names or a recording of a call. The person does not sort these into a corporate bucket and a customer bucket. They see one task that has to be done. If the session only says to avoid sensitive data, they will decide what sensitive means, and they will decide it while the meeting is starting.
AI governance training has to name both, in the language of the job. What in this team's week counts as company data. What counts as client data. Which of it can go into an approved tool, which of it has to be stripped first and which of it stays out. A slogan dies on a long pack. Real files from their week survive the walk back to the desk.
The tool question sits next to the data question. People bring their own tools because the approved one is slow, missing or unknown. Training that pretends the personal account is absent will be ignored by the people who already use one. The session has to say what happens when the allowed tool is worse than the one on the laptop. Which work can move. Which work waits. Who can approve an exception before the end of the day.
The check before anything leaves
A draft that stays on the screen and an output that goes to a client are different acts. Training that stops at a warning leaves both in the same bucket. The team needs a check with an owner. What a person reads. What they look for. What they do when the answer is fluent and wrong. The check has to be short enough to happen on a Thursday, or the deadline will win and the check will be the thing that gets skipped.
The question has to be cheap as well. When people are reluctant to admit the important use, a town hall collects the safe examples. Name a person who answers this week. Write the answer where the next person can find it without booking time. A question that vanishes into a mailbox teaches the room to stop asking. When the page exists and nobody can locate it, that problem is already set out in you have an AI policy and almost nobody can find it.
I ask a sponsor for one live case before we book. A file someone was about to drop into a tool, or already did. The name of the person who owns the rule. The name of the person who does the work. If those three are missing, the day becomes a tour of a framework, and the following Monday looks like the last one.
A line for the file in front of them
The day has a shape that is easy to skip and expensive to skip. Put three real files on the table, including one the sponsor would rather not discuss. For each file the room writes one sentence: this can go in, this has to be stripped or this stays out, and the reason in plain words. Then they do the sort. The person who owns the work reads the sentence back in their own words. If they cannot, the sentence is still a committee line, and the five o'clock decision is still theirs alone.
Then run the awkward version. A senior person wants the client pack summarised before a call that starts in twenty minutes. The approved tool will not take the pack as it stands. The personal account will. The room practises the reply: what can be summarised from notes that carry no client names, what waits until after the call and who is allowed to say so. AI governance training that never stages this moment produces polite agreement and the same paste on Thursday. The senior person has to be in the room. A reply rehearsed for an empty chair will not survive the call.
What we practise in the room
Collective Campus builds high-performance teams for the AI world. The AI governance workshop is where that case gets a line the desk can use. Leaders and the people who touch the data sit together. We sort real files into what can go in, what must be stripped and what stays out. We practise the refusal when a senior person wants the client pack summarised now. We write the check, the time it takes and the name of the person who owns an exception. What a product or operations team needs before an agent is allowed to act is in AI governance for non-lawyers.
The useful test is the following week, on a file that was not in the room. The person can point at the line. The tool they used is an allowed one, or an exception someone named. A human has read anything that left the building. If none of those is true, the workshop was a discussion. AI governance training earns its place when the next risky file is handled without a fresh debate.
Before you lock the date
- Which file, from a real week, was someone tempted to put into a tool?
- What in that file is company data, and what belongs to a client?
- Which tools are actually in use, including the ones you did not issue?
- Who answers the unclear case this week, and where does the answer get written?
Send me the case, the data it holds and the name of the person who can set the line. We will shape the day around that decision. A rule the desk has not practised is a document. The file is already in the window.
