Set the Ground Rules: Staff AI Policy, Privacy and Approved Tools

All lessons Lesson 2 of 7 • about 15 minutes A clinic manager and two staff gather at a reception counter as one of them initials a single sheet of paper.

A clinic manager and two staff gather at a reception counter as one of them initials a single sheet of paper.

By the end of this lesson: Write a one-page staff AI policy with a named privacy contact, a never-enter list and a short table of approved tools, each checked against eight privacy and cost questions.

Lessons 3 to 6 all have you putting business material into AI tools. Before that starts, your business needs one page of rules: which tools are approved, what never goes into them, and who checks the output.

The beginner course gave you a personal traffic-light rule for what to type. This lesson turns it into a rule for everyone who works for you, with a named privacy contact and a short table of approved tools. It becomes page 2 of your AI playbook.

Why the rules come before the tools

Canada's federal private-sector privacy law, PIPEDA, applies to organizations that collect, use or disclose personal information in the course of commercial activity. Alberta, British Columbia and Quebec have their own private-sector laws that generally apply instead for business inside the province, though PIPEDA still covers personal information that crosses a border. Ontario has a separate health-information law, PHIPA.

What matters for AI: the Office of the Privacy Commissioner of Canada (OPC) says a business stays responsible for personal information it transfers to a third party for processing, and must use a contract or other means to get comparable protection. An AI vendor is a third party. The usual leak is not a hack. It is a well-meaning employee pasting a customer's details into a personal free account nobody at the business chose.

Bill C-36, which would enact a new federal privacy law, was introduced on 15 June 2026. On 6 October 2026, when this lesson was checked, it was at second reading in the House of Commons and had not passed, so this lesson describes the law in force.

Four privacy moves for a small business

List what you hold. Write down the kinds of personal information you keep (contact details, addresses, payment records, appointment or health notes) and which software touches each.
Name a privacy contact. PIPEDA's accountability principle requires someone to be responsible for compliance, and its openness principle expects that person's name or title and contact information to be available. Usually that is the owner.
Update your notice. OPC guidelines expect you to tell customers in clear language that their information may be processed in another country and may be accessible to law enforcement and national security authorities there. Add a line to your privacy notice or intake form saying you use service providers, including AI tools, that may do this.
Know the breach steps. Contain it, write down what happened, and assess whether there is a "real risk of significant harm". If there is, PIPEDA requires a report to the Privacy Commissioner and notice to the people affected as soon as feasible. Keep a record of every breach for 24 months. Knowingly ignoring these duties is an offence, with fines of up to $100,000. OPC guidance treats the business that controls the information as responsible for reporting even when the breach happens at its processor.

Staff records differ: PIPEDA covers employee information only in federally regulated businesses such as banks, so in most Ontario shops keeping staff files out of AI tools is good practice rather than a PIPEDA duty. Questions go to the OPC Information Centre at 1-800-282-1376, Monday to Friday, 9 am to 4 pm Eastern. This is general information, not legal advice.

Plans and settings matter more than the brand

The same assistant can be safe or unsafe depending on the plan and its switches. From the vendors' own pages on 6 October 2026:

Personal plans. For the personal Gemini app, Google says a subset of chats is read by human reviewers, reviewed chats can be kept for up to three years, and users should not enter confidential information.
Business plans. Google says Workspace content is not used to train models outside your domain without permission. Anthropic says inputs and outputs from its commercial products, such as Claude for Work, are not used for training by default unless you send feedback or opt in (its Team plan needs at least two members). Microsoft says Copilot prompts and responses used by organizations under its commercial terms are not used to train foundation models.

So the policy says business accounts only, one login each, two-step sign-in. The Canadian Centre for Cyber Security's generative AI guidance recommends setting policies on how AI is used, multi-factor authentication, and keeping personal or sensitive business data out of prompts.

Choose in this order. First, AI features inside software you already pay for (email, accounting, booking, point of sale). For example, on the same day Google's Canadian pricing page listed its Workspace Standard plan at CA$18.40 per user per month on monthly billing, before tax and before introductory or annual-commitment discounts, with Gemini in Gmail, Docs and Meet. Prices change, so check the page yourself. Second, one general assistant. Specialized tools last. Cancel anything that overlaps.

Example arithmetic, replace the numbers with your own: $35 a seat x 3 staff x 12 months = $1,260 a year. At an example cost of $30 an hour, that tool must save 42 hours a year, just under 50 minutes a week across the whole team, to pay for itself.

Eight questions for any tool or vendor

Answer these from the vendor's own pricing and privacy pages, not from a sales call.

Is our data used to train the model?
Where is it stored, and who can see it?
Can I add and remove staff and see usage?
What is the all-in price, including overage, and in which currency?
Is it month to month?
Can I export everything when I leave?
Is there support I can reach?
Can I speak to a current customer like me?

Red flags in a pitch: guaranteed leads or rankings, no published prices, long lock-in, "set and forget" claims, a guaranteed grant, or terms under which the vendor owns your phone number, ad account or website. On grants, BDC's page says the Canada Digital Adoption Program stopped accepting new applications for its Boost Your Business Technology stream on 19 February 2024, so check any funding claim at the source.

The seven lines of a workable staff policy

Approved tools only, on business accounts, with one login each and two-step sign-in.
The never-enter list, in your own plain words.
A named person reviews anything before it reaches a customer or the books.
Honesty: say when someone is dealing with an automated system, and no fake reviews, testimonials or results.
No AI-only decisions about people.
Marketing messages pass the Lesson 4 anti-spam check.
Mistakes are reported fast and without blame, and here is who to tell.

Line 4 reflects the generative AI principles Canada's federal, provincial and territorial privacy commissioners published in December 2023: people should know when they are dealing with a generative AI tool.

Some readers have extra rules. Ontario clinics handle health information under PHIPA. Quebec's private-sector law, updated by Law 25, adds duties such as a privacy impact assessment before personal information is sent outside Quebec. Regulated professionals should ask their regulator before using AI with client files. Review the page every six months; tools, prices and the law change. This is general information, not legal advice.

Try it: Write and sign your one-page staff AI policy (about 15 minutes for the draft, of which about 10 is in your AI assistant; initials and vendor checks can finish this week)

Ask every staff member, with no blame attached, which AI tools they already use for work, and add your own. List every AI or software subscription you pay for.
Pick the one or two tools you will approve. Run each through the eight questions using the vendor's own pricing and privacy pages, then mark it approved, approved with limits or not approved. If you find two subscriptions doing the same job, cancel one.
Write your never-enter list in your own plain words. For example: customer names with health or payment details, card numbers, social insurance numbers, passwords, staff records, anything told to you in confidence.
Give your AI assistant the seven lines, your approved-tools table and your never-enter list (rules only, no real customer or staff details) and ask for a one-page policy. A free AI chat app is fine for this step because nothing confidential goes in. If you wrote a one-page rule in the beginner course, start from that. Rewrite anything that does not sound like your shop.
Add the privacy contact's name and the 'if something goes wrong' step: who to tell the same day, and that this person contains it, writes down what happened and assesses the risk.
In each approved tool, set the training and history options to match the policy and turn on two-step sign-in. Lesson 6 of the beginner course covers changing the key privacy settings; if you cannot find a switch, search the vendor's own help pages.
Date the page, have each staff member read and initial it (a solo owner signs it too), and put a six-month review in your calendar. Save it as page 2 of your AI playbook.

Prompts you can copy

Ask staff what they already use

Write a short, friendly message to my [number] staff at [business name]. Ask which AI tools they already use for work, whether each is on a personal or a business account, and which tasks they use it for. Make it clear nobody is in trouble and that I am asking so we can agree on approved tools and simple rules. Ask for replies by [date]. Keep it under 90 words, in a [tone] tone.

Check a vendor against the eight questions

I am deciding whether to approve [tool name and plan] for my business. Below is text I copied today from the vendor's own pricing page and privacy page: [paste the page text]. Using only that text, answer these eight questions: 1) Is our data used to train the model? 2) Where is it stored and who can see it? 3) Can I add and remove staff and see usage? 4) What is the all-in price, including overage, and in which currency? 5) Is it month to month? 6) Can I export everything when I leave? 7) Is there support I can reach? 8) Can I speak to a current customer like me? For each answer, quote the sentence you relied on. If the text does not answer a question, write 'not stated' and do not guess. Finish with a list of questions I should send the vendor.

Draft the one-page staff AI policy

Write a one-page staff AI policy for [type of business] with [number] staff in [city, province]. Use plain language a new hire can read in two minutes. Build it from these seven rules: [paste your seven lines]. Approved tools table: [tool, plan, what it may be used for, any limits]. Never-enter list: [your list]. Privacy contact: [name or job title]. If something goes wrong, tell [name] the same day. Add a line for each staff member's initials and the date, and a review date of [date six months from now]. Do not add any rule I did not give you. At the end, list anything unclear or missing as questions for me.

Privacy notice line, for checking before you publish

Draft two plain-language sentences for the privacy notice and intake form of [type of business] in [province]. They should say that we use outside service providers, including AI tools, to [purposes, for example draft replies and summarize notes], that these providers may process information outside Canada, where it may be accessible to law enforcement and national security authorities in that country, and that people can contact [privacy contact name or title] at [business email or phone] with questions. Give me two versions, one shorter than the other. This is a draft only; I will have the wording checked before I publish it.

Before you move on

I asked staff which AI tools they already use and listed every subscription we pay for. Each approved tool passed the eight questions from the vendor's own pages, and any overlapping subscription is cancelled. Approved tools are on business accounts with one login per person, two-step sign-in on, and training and history settings matched to the policy. My never-enter list is written in my own words and every staff member has read it. A privacy contact is named, the policy says who to tell when something goes wrong, and our privacy notice mentions service providers outside Canada. The policy is dated and initialled, saved as page 2 of the AI playbook, with a six-month review in the calendar.

Remember: Your business answers for customer information wherever it goes, so put the approved tools, the never-enter list and the named reviewer on one signed page before anyone uses AI on customer work.