# Your Staff Are Already Using AI. Start With a Register.

> Most staff using AI at work brought their own tools, and many keep it quiet. Before any policy, write down what is already running. Here is how to build an AI register in a week.

By Warwick Heathwood. Published 2026-10-23; facts last checked 2026-10-23. Canonical: https://subfractal.com.au/thinking/start-with-an-ai-register

## The short version

- Your people are almost certainly using AI tools you didn't choose, and some are keeping it quiet. Banning them rarely works. The first useful step is a register: a plain list of which AI tools are in use, who uses them, for what, what information goes in, and who checks what comes out. Everything else in AI governance is built on that list.

Greg Isenberg spent 48 hours this month with AI founders in San Francisco and came home with thirty-one observations. One of them will be familiar to anyone who runs a business: people everywhere are building their own little tools for their own jobs. In his words, it's "amazing for productivity and a nightmare for IT, because nobody knows what's running or who owns it."

That isn't only a Silicon Valley problem. It's probably happening in your office this week.

## How many staff use AI without telling anyone?

More than most owners expect. In Microsoft and LinkedIn's 2024 Work Trend Index, a survey of about 31,000 knowledge workers in 31 countries, 78% of people using AI at work were bringing their own tools rather than using ones the company provided. At small and medium businesses the figure was 80%.

People are also quiet about it. The University of Melbourne and KPMG's 2025 global study of trust in AI surveyed more than 48,000 people in 47 countries. It found that:

- 57% of employees hide their use of AI and present AI-generated work as their own;
- almost half admit to using AI in ways that go against company policy, including putting sensitive company information into free public tools;
- only 40% say their workplace has any policy or guidance on generative AI.

So the usual picture inside a small firm looks like this. Several people use ChatGPT, Claude, Gemini or Copilot on their own accounts. One or two have built something cleverer: a spreadsheet that calls an AI, or a browser extension that writes their emails. Nobody has written any of it down.

## Why doesn't banning AI work?

Because the work it saves is real, and people don't give that up because a memo says so. A ban mostly moves the use to personal phones and accounts, where you can see even less. It also tells your most curious people that their curiosity is a problem.

The better question is not "should we allow AI?" but "what is already running, and what should it be allowed to touch?" You can't answer the second without the first.

## What is an AI register?

An AI register is a list of every AI tool or AI-assisted process in use in your business, with enough detail to make decisions about each one. It is the AI equivalent of an asset register or a list of approved suppliers.

You don't need to invent one. Australia's National AI Centre publishes a free AI register template alongside its Guidance for AI Adoption, released in October 2025. It records each system's name, owner, status and source, and the guidance treats knowing what you use as a foundation for everything else.

For a business of 15 to 200 people, a register that fits on one page is a good start.

| Column | What to write |
|---|---|
| Tool | The product and plan: "ChatGPT, personal free account", "Copilot in Outlook", "the quoting spreadsheet Sam built" |
| Who uses it | Names or roles |
| What for | The actual task: drafting quotes, summarising reports, writing job ads |
| What goes in | The kind of information: public, internal, client-confidential, personal |
| Who checks what comes out | The named person who reviews the output before it's used |
| Decision | Allow, allow with limits, or stop |
| Review date | When someone looks at this row again |

## How do you build an AI register in a week?

1. **Ask, don't audit.** Send one message: "We want to know which AI tools you're using so we can support them properly. Nobody is in trouble." An amnesty gets you far more truth than a software scan.
2. **List the tools and what they touch.** For each one, record what information goes in. This is the column that matters most. A free public tool fed client reports is a different risk from a paid tool used to tidy a job ad.
3. **Name an owner for each row.** The person who would answer for it if a client asked.
4. **Make a decision per row.** Most rows end up as "allow". Some become "allow, but never with client information", or "use the paid business account, not the free one". A few become "stop".
5. **Put a date on the review.** Tools change monthly. Look at the register every quarter, and whenever someone builds something new.

That's a week of a manager's spare hours, not a project.

## What does the register let you do next?

**Write rules that fit.** An AI policy written before you know what's in use is guesswork. One written from the register describes your real business: which tools, which information, who signs.

**Answer the questions that are starting to arrive.** Clients ask in tenders. Insurers have started asking. If you are covered by the Privacy Act, from 10 December 2026 your privacy policy has to explain certain decisions that a computer program helps make, where they significantly affect people. You can't describe what you haven't listed. If you're an accredited facility, your assessor will want to know what technology touches your results.

**Back the good ideas.** The register usually turns up one or two homemade tools that are genuinely good. Those are the ones worth making properly, with the checks they need, instead of leaving them on someone's personal account.

## What a register is not

It isn't a surveillance exercise, and it isn't a compliance certificate. It is the honest starting point for deciding what AI may do in your business. It works best when the people using the tools help write it.

## Sources

- [Greg Isenberg, "48 hours in SF", Greg's Letter, 22 October 2026 (observation 20, personal software)](https://www.gregisenberg.com/)
- [Microsoft and LinkedIn, 2024 Work Trend Index: AI at work is here. Now comes the hard part](https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part)
- [University of Melbourne and KPMG, Trust, attitudes and use of artificial intelligence: a global study 2025, and its media release](https://fbe.unimelb.edu.au/newsroom/media-release-global-study-reveals-trust-of-ai-remains-a-critical-challenge-reflecting-tension-between-benefits-and-risks)
- [National AI Centre, Guidance for AI Adoption (October 2025) and AI register template](https://industry.gov.au/publications/guidance-for-ai-adoption)
- [Privacy Act automated decision-making transparency, commencing 10 December 2026: Gilbert + Tobin summary](https://www.gtlaw.com.au/insights/automated-decision-making-transparency-under-the-privacy-act)
