Back to the blog

Govern AI by what you build

Privacy practice works by blocking because an organization owns the places its data lives. AI use happens in a text box it does not own. The control that works there is a better path that keeps a record.

Where We Left Off

Last time we looked at AI for All, Canada's national AI strategy. It sets an adoption target with a year attached, 60 percent of businesses by 2034, and it rests that target on a federal privacy law in force since 2000. We wrote that a business working under that law already knows the shape of the obligation ahead of it.

That last line is the one we kept arguing about afterwards. The law carries over. The question is how much of the practice built around it carries over too, and the longer we looked, the more it came apart into two halves.

What Privacy Practice Is Good At

Privacy practice starts from the data. An organization maps what it holds, classifies it, limits who can reach it and deletes it on a schedule. The techniques are mature. Tokenizing an identifier, swapping a name or an account number for a stand-in that means nothing outside the system, has been standard work for many years.

All of it rests on one condition that rarely gets said out loud. The organization owns the places the data lives: its servers, its applications and the laptops it hands out. Ownership is what makes the strongest tool in the kit possible, and that tool is the block. Close the port. Lock the folder. Wipe the laptop when someone leaves.

The Place Moved

AI tools reached people as consumer products, and that changes the condition.

Picture a clerk with forty resident complaints to answer before the end of the day. A chat window in the browser drafts a decent reply to each one in seconds. All it asks is that the complaint be pasted in, and the complaint has a name, an address and a phone number in it.

Nothing in that story touches a server the organization owns. The text box belongs to someone else. One person decides what goes into it, at their desk, for the task in front of them, and the data classification written two floors up never enters the picture.

The shortcut also pays at once. That is what makes it hard to forbid. The clerk goes home on time. A rule that says no competes with that reward every working day, and people work around a block that stands between them and a faster way to do their job.

Build the Path

So the control that works here is a path. Give people a tool that does the job better than the one they would find alone, and the choice makes itself. Nobody takes on risk or cost for a worse result.

From the building side the design is plain. Make one gate that the work flows through naturally, and put the controls in the gate.

We build software, so here is what that looks like in ours. Every AI request in the product, from any screen, leaves through one doorway in the code. Before the request leaves, the doorway writes a line to the audit record: who asked, what for, on which record, and which AI service answered. On the same pass it replaces the names of the people in the directory with tokens, along with the emails and phone numbers it finds, and it puts them back when the answer returns. The customer's administrator chooses which AI service sits on the far side, and one they host themselves is among the choices.

For the clerk, the governed tool is the one already open on the screen, inside the record they are working on. It is the shortest path to the reply. That is the whole design.

It lowers the risk. Some risk remains, and the audit record shows where.

What Changes for Governance

The governance conversation changes with it. The old question was what to block. The new one is what to build: which tool people get, what it records, what it takes out before anything leaves and who reads the record afterwards.

A policy then has something to point at. "Use the approved tool" is a rule people follow when the approved tool is the easy way to work. It is a rule they resent when it is the slow way.

The rule your own jurisdiction wrote still decides what the gate must do. The building is how an organization meets it.

Summary

  • Privacy practice works by blocking because the organization owns the places its data lives.
  • AI tools reached people as consumer products. A clerk pastes a complaint into a browser, and nothing in that touches a server the organization owns.
  • The shortcut pays at once, and people work around a block that stands between them and a faster way to work.
  • The control that works is a path. It is a tool that does the job better, and every request passes one gate.
  • In our own product the gate records who asked and what for, swaps names, emails and phone numbers for tokens, and leaves the choice of AI service to the customer.
  • Governance moves from what to block to what to build. Your jurisdiction's rule decides what the gate must do.

Ask for a demo

Contact us. We walk through Aptli on the record types you name.