The AI Feature That Turned Itself On: Deciding What Ships to Staff

Monday's update to the office suite adds an AI assistant, switched on by default. Nobody bought it. Nobody reviewed it. It is simply there, summarizing documents and drafting replies for anyone who clicks it.

The question is not whether the feature is impressive. It is whether anyone in your company decided you should use it. An unmade decision is still a decision; it is just one the vendor made for you.

Key takeaways

• Default-on AI features bypass whatever procurement review you think you have.

• The real questions are what data the feature reads and where its output goes.

• Someone with authority must own the keep-it-on or turn-it-off call for each feature.

• Off until reviewed is a legitimate default for sensitive teams and sensitive data.

• Record each decision in the same inventory as the tools you meant to buy.

Step 1: Notice what the update actually changed

Vendors announce these features in release notes almost nobody reads. The fix is a habit, not a project: whoever administers each major suite scans the notes for new AI capabilities and flags them to the person who owns the decision.

Five minutes per update cycle is usually enough to catch the moment a tool your staff already uses quietly becomes a different tool.

Step 2: Ask the three questions

For each new feature, three questions carry most of the weight. What data does it read: the document in front of the user, or everything the account can reach? Where do prompts and outputs go: do they stay in your tenant, or are they used to improve the vendor's models? Can it only suggest, or can it act?

The vendor's documentation and admin console answer these more often than you would expect. When they do not, that silence is itself an answer about how ready the feature is for sensitive work.

Step 3: Decide by audience, not by feature

Rarely is the answer a flat yes or no for the whole company. Drafting help for general staff may be fine while the same feature stays off for the people handling customer records, personnel files or board materials until it has had a real review.

This fictional example shows the decision record. It is an illustration, not a prescription.

The table is the deliverable: one line per feature, a name and a date on each.

Step 4: Write the decision where people can find it

A decision that lives in one person's head expires when they change jobs. The AI inventory and the one-page policy exist so the answer outlives the conversation. Add the feature, the decision and the review date, and tell staff what changed in the same breath.

Step 5: Set the review, then get back to work

These features change quarterly. A decision made in November deserves a look in February: did the vendor's data handling change, did staff find a use you did not expect, did the regulator chatter become an actual rule?

The shadow AI approval inventory covers finding the tools nobody bought, and the one-page AI policy is the place these decisions get written down for staff.

Frequently asked questions

Is turning a feature off an overreaction?

Not when it is a pause rather than a ban. Off until reviewed says the company decides what reads its data, which is the same standard applied to any new vendor. Features that pass review get switched on with a clear record.

Who should own these decisions?

Whoever owns the data the feature touches, advised by IT. For company-wide tools that is usually an executive sponsor; for team tools it can sit a level down. What matters is that the owner is named and knows they own it.

What if a feature cannot be disabled?

Then the decision moves up a level: does this product still belong in the environment, and with what handling rules for sensitive material? A vendor that removes your ability to say no has made the decision for you, which is worth weighing at renewal.

How is this different from the AI inventory work?

The inventory finds the tools. This is the ongoing habit of catching the tools that arrive inside tools you already approved. Both end up in the same register.

A Defensible Next Step

Pick your most-used suite and list which AI features are on today. If nobody can produce the list, that is the finding. Book a Decision-Clarity Call if you want help setting up the review habit.