Your people are already using AI, and a policy that says otherwise only changes where they use it. The decision in front of most leadership teams is not whether AI enters the business. It is which account it enters through.
I ran this research for my own organization, and the answer surprised people in a useful way. The gap between a personal subscription and an enterprise agreement is not about capability. Both tools write the memo. The gap is about what happens to the memo afterward.
What actually differs between a personal account and an enterprise one
Three things, and none of them are visible to the person typing.
- Training. Consumer accounts, including paid personal subscriptions, may use what is typed into them to improve future models. Enterprise agreements generally do not, and that difference is contractual rather than a setting.
- Retention and location. An enterprise tenant can tell you where the data sits, how long it is kept, and who at the vendor can reach it. A personal account cannot answer any of those questions on your behalf.
- Administrative visibility. With a tenant, you can see adoption, apply conditional access, and revoke a departing employee in the same motion as their email. Without one, an ex-employee keeps whatever they pasted in.
There is a fourth difference that matters more than the first three combined. A tool inside your tenant can be given access to your own documents under your own permissions, which means it gets better at your business rather than better at the internet.
Why people reach for the consumer tool anyway
This one catches all of us. The consumer tool is one click away, it does not require a request, and it works. The person pasting a contract into it is not being careless. They are trying to finish something on a deadline with the fastest tool in front of them, which is the same instinct that makes them good at the job.
That is worth saying plainly inside the business, because the alternative framing, where employees are treated as the risk, produces quieter use rather than less of it. The honest version is that the company did not give them a sanctioned option fast enough.
The four decisions a leadership team owns
Governance sounds heavy. In practice it is four decisions, and they can be made in a single meeting.
- Which tools are sanctioned. Name one or two and provision them properly. A short list that people can actually get access to beats a long list nobody has been granted.
- What may never be typed in, regardless of tool. Customer personal information, credentials, unreleased financials, anything under an NDA. Keep the list short enough that people remember it without looking.
- Who decides on a new tool. A named person and a turnaround commitment. If a request takes six weeks, the answer will be a personal account.
- What gets reviewed, and when. Adoption, incidents and the sanctioned list, once a quarter. This market moves faster than an annual policy cycle.
Notice what is not on that list. There is no model evaluation framework and no committee. Those may come later, at a scale that justifies them, and starting there is how AI governance efforts stall before they protect anything.
What a usable acceptable-use policy contains
One page. The sanctioned tools, the never-type list, the request path with a named owner, and a plain statement that output gets checked by a human before it leaves the building. That last clause does more work than the rest of the policy combined, because most AI incidents in normal companies are accuracy failures rather than data leaks.
If the policy runs longer than a page, it will be read once during onboarding and never again, and a policy nobody remembers is a control that does not exist.
What to do on Monday
Find out what is already in use. Ask directly rather than auditing, because people answer honestly when the question is framed as provisioning rather than enforcement. You will usually find three or four tools in regular use and one that gives you pause.
Then sanction the option you already pay for. Most organizations running Microsoft 365 or Google Workspace already have an enterprise-grade assistant included, sitting unprovisioned, while people pay personally for a consumer one. Closing that gap costs nothing and removes most of the exposure in a week.
If you are working out where your organization stands on this, an AI governance review produces the sanctioned list, the one-page policy and the oversight cadence in a fixed scope. Tell me which tools are already in use and I will tell you which of them are the real problem.