Count what runs, not what's chartered. Assistants staff bring, department builds and vendor features all belong in the count.
Make the list before the strategy. Readiness and risks come after an inventory of what actually runs.
Hold every unit accountable. A central office cannot see the AI a department built for itself, so every unit names its own finder.
Watch without spying. Publish what is monitored and which sites, and coach in the moment instead of punishing.
Ask openly, and test your policy. If admitting unapproved use feels riskier than hiding it, the policy has already failed.
A free lesson from Governance, risk and responsible AI, one of its 26 lessons. Read another free lesson
When the approved list stops working
In Gallup's 15 June 2025 report, 40% of US employees said they had used AI in their role a few times a year or more. That was nearly double the share in 2023.
Companies typically ran three stages: block generative AI, form a council, then selectively enable approved apps. By February 2025 that third stage was breaking, because AI features were appearing inside every SaaS tool already in use. A council that charters a short list of approved apps is governing by hopes and prayers. The risk lives in the gap between the sanctioned list and what actually runs. Staff cross it to get the work done, not out of malice. That gap is what the inventory is for.
What the count has to include
Start with an inventory of every AI system in use. You cannot manage what you cannot see. Readiness and risks come after the list.
The assistant in the browser tab. Deloitte UK's inaugural GenAI Workforce Survey, published 16 September 2026, found 31% of UK GenAI users using it without their employer's knowledge. The first risk is compliance. Well-meaning staff paste regulated data, like patient information, into an unsanctioned assistant to be productive, breaking rules like health privacy law.
The agents staff wire up themselves. A team member can connect an assistant to the shared inbox and have it draft replies. No council chartered it. Ask openly what runs on team accounts.
The AI a department built itself. Even with oversight from the CIO's office, departments go off and build their own AI without anyone knowing. Make every unit accountable for naming what it runs, not just the central office.
The AI inside software you already bought. By September 2025, nearly every third-party application and ERP vendor advertised AI features. That adds third-party risk on top of the AI you build, so it belongs in the inventory and in the risk assessment.
How the watching itself is run. Say plainly what is monitored and which sites. Involve the person being watched. Most leaks are accidental oversharing, so coach in the moment instead of punishing. Before you reach for detection or the disciplinary policy, ask staff which assistants and agents they already run. The founder of an HR training firm argues for that open ask, testing whether an admission makes the next person likelier to speak up. If admitting it feels riskier than hiding it, "your policy has already failed." Hold your own policy to that test. Tracking each developer's AI token use is legitimate cost management, not surveillance, when three conditions hold. It must protect privacy, limit what data is collected, and use the data only for its stated purpose.
The inventory on one page
Keep it as a running list with a named finder on every row:
Where AI hides
Who finds it
How often
Assistants staff bring themselves
The staff using them, asked openly
A date you set
Agents wired to staff accounts
Each team, in the same open ask
A date you set
AI a department built itself
The unit that built it
A date you set
AI inside software you already bought
You, through your risk assessments
A date you set
The monitoring itself
You, with the people it watches
A date you set
The dates are yours to set. Any row without a named finder and a date is the next thing to fix.
Exercise (20 minutes): fill in the AI inventory card
Run this on your own organization, not on a hypothetical one. Fill in what you know from where you sit; you won't know every line, and the lines you can't fill tell you who you're missing. Every line you can't fill gets an owner and a date before you close the card. The card download includes a filled-in example of an organization that isn't ready yet, and why.
AI inventory card (PDF)
A blank card and a filled-in example, one page each.
These three prompts go into your company's approved AI assistant. Use the first to fill in the card one row at a time; it won't guess an answer for you. Use the second when the card is done, to hear the case against calling the count finished. The third asks about your industry and your organization and turns the lesson into a 30-day plan, with a checklist, rubric or brief for your boss if you want one. Use placeholders for company, system and people names; the help doesn't depend on them.
Prompt 1: Fill in my card with me
Paste this, then add your organization in one or two sentences and the card below.
Prompt
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X) and round any figures. The help you get does not depend on them.
You are helping me fill in an AI inventory card for my organization: a count of every AI system already in use, plus how the watching itself is run. The card is pasted at the end.
Rules:
- Ask me about one row at a time, in order, and wait for my answer before moving on.
- Write down only what I tell you. Never suggest, estimate or invent a number, a name, a yes/no answer or a decision for me, even if I ask you to guess. If I ask, say that this row needs a real answer and ask who could give it.
- If I don't know an answer, write "Not known" and ask who should find out and by when. Put that in the Owner / by when column.
- After the last row, show the whole card as a table with three columns: Question, Answer, Owner / by when. Then list every row marked "Not known" with its owner, and remind me that any unanswered row means the count is not done: that row is the next thing to fix.
My organization in one or two sentences: [DESCRIBE YOUR ORGANIZATION AND WHERE YOU ARE STARTING]
The card:
[PASTE THE CARD]
Prompt 2: Stress-test my card
Paste this, then your filled card.
Prompt
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X) and round any figures. The review does not depend on them.
Below is an AI inventory card I filled in for my organization. Act as a skeptical reviewer at the decision meeting and argue the case for NOT READY: the count is not finished.
For each row:
- Say whether the answer is specific enough to act on (a number, a named role, a date) or too vague.
- Point out where answers contradict each other, for example what the monitoring covers against what staff have been told is published, or an open ask promised in one row against a disciplinary step in another.
- Say what evidence you would ask for before accepting it.
Do not rewrite my answers and do not invent facts about my organization. If a row is blank or says "Not known", treat it as not ready.
Finish with: your recommendation (INVENTORY READY or NOT READY), the three rows that most need work, and one question I should ask my team on Monday.
My card:
[PASTE YOUR FILLED CARD]
To paste the card into prompt 1, copy its rows from the PDF.
Prompt 3: Make it yours
Paste this and answer its four questions.
Prompt
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X, Supplier B).
You are helping a leader apply one lesson to their own situation. The lesson's five decisions are below. Use only them and what I tell you. Never invent facts, numbers, vendors, laws or results about my organization or my industry. Where a number is needed and I have not given one, say what to measure and who should measure it, instead of guessing.
First, ask me these four questions, one at a time, and wait for each answer:
1. My industry, and roughly how many people work in my organization.
2. What I am counting first, in one or two sentences: the whole organization, one division, or one team.
3. Where my count stands today: nothing started, an approved-apps list only, or a partial inventory.
4. What worries me most about it.
Then give me the following. Keep it short and practical:
A. My plan. For each of the five decisions: one action written for my situation, the role that should own it (a role, not a name), and what should be done within 30 days. If my industry makes a decision harder, for example regulated data, client confidentiality or staff working away from company devices, say which decision and why in one sentence.
B. Three questions to ask my team on Monday, specific to my count.
C. Ask me which one I want: a checklist for my count, a rubric to score whether a row is really settled, or a one-page brief for my boss. Then write it, and finish with a prompt I can paste later to update it.
Mark anything that rests on an assumption with "(check)".
The five decisions:
1. Count what runs, not what's chartered: assistants staff bring themselves, agents teams wire up on their own accounts, AI a department built itself and AI inside software you already bought all belong in the inventory.
2. Make the list before the strategy: the inventory of what actually runs comes first, and readiness and risks come after it.
3. Hold every unit accountable: a central office cannot see the AI a department built for itself, so every unit names its own finder, with a date.
4. Watch without spying: publish plainly what is monitored and which sites, involve the people being watched, and coach accidental oversharing in the moment instead of punishing.
5. Ask openly, and test your policy: before detection or discipline, ask staff which assistants and agents they already run. If admitting unapproved use feels riskier than hiding it, the policy has already failed. Any tracking of AI use, such as developers' token use, must protect privacy, limit what data is collected and use it only for its stated purpose.
Your company's assistant may answer differently, so read what it writes before you use it.
Check yourself
Your approved list has ten apps on it. Is that your inventory?
Show the answer
No. Staff assistants, department builds and vendor features sit outside it.
A department built its own assistant; the CIO's office never saw it. Who owns the fix?
Show the answer
That department's head; a central office alone cannot see these.
Your ERP vendor ships an AI feature in the next release. Where does it go?
Show the answer
Into the inventory and the risk assessment: third-party risk on top of what you build.
You want visibility into staff AI use without spying on them. What do you publish first?
Show the answer
What is monitored and which sites, then coaching in the moment instead of punishment.
Security proposes tracking developers' AI token use. When is that cost management, not surveillance?
Show the answer
When it protects privacy, limits what data is collected, and uses it only for the stated purpose.
Sources
The advice draws on interviews with a startup founder, the COO of a market-intelligence firm, the CEO of a technology services firm, the CTO of a data-security startup and the head of AI at a security and observability data platform.
On the numbers: the Gallup figures are self-reported and cover US employees; the "did not know" response was new in the December 2025 wave, so it is not directly comparable with earlier waves, and Gallup's link from that gap to personal tools is an inference. Deloitte's survey is UK workers' own reports and its first wave; "without their employer's knowledge" is the worker's belief, not an audit.