Who Approved the AI Tool Your Team Already Uses?

Somewhere in your company, a proposal was drafted this week with the help of an AI tool nobody approved. The draft was good. The data in it - pricing, a customer name, an HR detail - now sits in a service your contracts do not cover.

This is not a discipline problem. It is an adoption lag: useful tools spread at the speed of a shared link, while approval processes move at the speed of quarterly reviews. The tools won.

The fix is not a ban. Bans push usage to personal devices where you cannot see it at all. The fix is an inventory, a fast decision on each tool, and an approved path people actually prefer.

TLDR

• Assume AI tools are already in use; the question is whether you can see which ones.

• Inventory from expense reports, SSO logs and team leads, not from a survey people answer politely.

• Decide each tool: approve with guardrails, restrict to safe data, or replace with a sanctioned option.

• The approved path has to be as easy as the shadow one, or the shadow one wins.

• Repeat the inventory quarterly; the tool set changes that fast.

Find what is actually in use

Polite surveys undercount. People either do not realize the browser plugin they installed is an AI service, or they suspect the honest answer ends with losing a tool they rely on. Better sources: expense reports and corporate card statements for subscriptions, single sign-on and browser telemetry where you have it, and team leads asked in private which tools make their people faster.

NIST's Generative AI Profile treats an inventory of AI uses as a starting point for governance. You cannot govern what you have not found.

Decide each tool once, in writing

For each tool found, make one of three calls. Approve it with guardrails: which data classes may go in, which account type to use, and whether outputs need review. Restrict it: allowed for public or internal data only. Or replace it: shut it down and give people a sanctioned alternative that does the job.

This fictional example shows a decision register. It is an illustration, not a real inventory.

Tool found	Used for	Decision	Guardrail
AI writing assistant	Proposals, emails	Approve	Company account; no customer data in prompts
Meeting transcription app	Sales calls	Restrict	Internal meetings only; consent check on calls
Code completion plugin	Engineering	Replace	Move to licensed enterprise version

The register matters more than any single decision. It is the difference between 'we have an AI policy' and 'we know what we approved.'

Make the approved path the easy path

People route around slow approvals. If getting an approved AI account takes six weeks and a procurement review, the personal account wins. The approved option needs to be provisioned in days, paid centrally, and good enough that nobody misses the shadow tool.

Watch the data flow, not just the tool. A sanctioned tool misused - pasting a customer list into the wrong prompt field - produces the same exposure as a banned one. Guardrails belong in onboarding and in the contract, not just in the policy PDF.

Keep the inventory alive

The tool set turns over quarterly. New entrants arrive, existing tools change their data terms, and last quarter's approved choice can become this quarter's laggard. Rerun the discovery sources on a schedule and revisit decisions when a vendor's terms change.

Assign the register an owner - usually whoever runs AI governance or the security program - and report movement: tools added, tools removed, and any data-class exceptions granted.

The related discussion of building trustworthy AI is worth your time alongside this one.

Frequently asked questions

Is a full ban ever the right answer?

Rarely, and mostly where regulatory constraints leave no room. A ban without an approved alternative drives usage to personal devices and personal accounts, which removes your visibility entirely.

Who should run the inventory?

Security or IT can run the technical discovery, but the decisions need the business: legal for contract terms, compliance for regulated data, and the team leads who know the workflows. Give the register one accountable owner.

What about AI features inside tools we already approved?

Treat them as new functionality. An approved vendor flipping on an AI feature that processes your data differently is a change in the arrangement, and it belongs in the register with its own decision.

A Defensible Next Step

Ask three team leads this week which AI tools actually make their people faster, and compare the answers to your approved list. The gap is your real AI posture. Book a Decision-Clarity Call if you want help building the register.