Most companies now use AI in more places than anyone has written down. Customers ask about it in security questionnaires, insurers ask about it at renewal, and the EU AI Act and privacy law apply to it. None of that needs a compliance department. It needs the ten steps below, done once and then kept up.
1. List every AI system
Include the AI you build into your product, the AI inside tools you buy (CRM, help desk, accounting, recruiting), and the assistants staff use on their own. The last group is usually the longest and the least known: ask, rather than assume.
2. Note what data goes into each one
For each system, write down whether it sees personal information, customer data, or confidential business data, and whether that data leaves your company to do so. This one column answers most later questions.
3. Sort them by risk
Use the EU AI Act’s levels even if you are not in Europe: banned, high risk, transparency, and everything else. Anything that helps decide about a person (hiring, credit, pricing, access to a service) needs the most care. Our article on the EU AI Act for Canadian and US companies explains the levels.
4. Write a one-page AI policy
Which tools are approved, what data may never be pasted into them, who approves a new tool, and what to do when something goes wrong. One page that people read beats twenty that nobody does.
5. Give each system an owner
One named person per system, who knows what it does, answers questions about it, and decides when it changes or goes.
6. Check your vendors
- Do they train their models on your data, and can you opt out?
- How long do they keep prompts and outputs, and where are they processed?
- Which sub-processors do they use?
- Can they show how they meet the EU AI Act or a standard such as ISO/IEC 42001?
7. Disclose what people should know
Tell people when they are talking to a bot, label AI-generated images, audio and video, and tell people when a decision about them was made automatically. Quebec’s Law 25 already requires the last one when the decision is based only on automated processing of personal information. App stores and marketplaces have their own AI disclosure forms; fill them in honestly.
8. Keep a human where it matters
For decisions that affect people, a person should be able to review the result, override it, and explain it. Write down where that check happens.
9. Keep records
For each system you build or configure: which model and version, what data it was trained or grounded on, how it was tested, and any incidents. When a customer or auditor asks, the answer is a document, not a meeting.
10. Train people, then review twice a year
Half an hour on the policy and on safe use of AI tools is enough for most staff. Then look at the inventory every six months: tools come and go faster than policies.
Frameworks worth borrowing from
You do not need to adopt a framework whole, but two are useful as checklists. The NIST AI Risk Management Framework organises the work into four functions: Govern, Map, Measure and Manage. ISO/IEC 42001 describes an AI management system and can be certified by an accredited body, which some enterprise customers have started to ask for.
If you would rather have help, our AI Governance & Compliance service covers all ten steps, with the technical side done for you and your lawyer kept in the loop.
General information, not legal advice.