← Guides·All compliance guides →
Your SituationWe are adding a new vendor or software →
Using AI in Your Business: A Privacy Guide for Canadian Small Businesses
By Yong DuPublished Reviewed
You can use AI and still meet your privacy duties. What regulators expect, and seven practical steps to take.
On this page
You can use AI in your business and still meet your privacy duties. Most of what regulators ask for is ordinary good practice: know what tools you use, be careful what goes into them, and have a person check what comes out.
No new law has to pass first. Canada's privacy commissioners said in 2023 that generative AI tools "do not occupy a space outside of current legislative frameworks". The privacy law that already applies to your business applies to the AI you use.
That's good news. You don't need a new compliance program, only a few clear decisions. This guide covers Alberta, BC and federal law. Quebec has its own rules and isn't covered here.
You are probably already using AI
In a 2024 BDC survey of 1,247 Canadian business owners, 39% believed they used AI. When shown a list of common tools, that rose to 66%. So 27% were using AI without realizing it, because it's built into software they already had.
It shows up in email, accounting, design, customer service and meeting software. The same survey found that 97% of smaller businesses using AI reported real benefits, and that data privacy was one of the barriers holding others back.
So the useful question isn't whether to use AI. It's whether you know where you already do, and whether those uses are ones you'd stand behind.
What the law expects today
The rules below come from existing privacy law, as Canada's regulators read it. Their 2023 principles are an interpretation, not a statute, but they say many of these points will be needed to comply with the law.
| The duty | What it means for AI |
|---|---|
| You stay accountable | Using a tool doesn't move responsibility to the vendor |
| Limited purpose | Use personal information only for purposes a reasonable person would find appropriate |
| Consent and openness | People should understand how their information is used, including with AI |
| Accuracy | Take reasonable steps to check output, especially for decisions about people |
| Safeguards | Protection must suit the sensitivity of the information |
One point surprises people: what an AI tool generates about an identifiable person is itself personal information, so the same rules apply to the output.
Seven practical steps
1. List the AI you use
Write down each tool, what it's used for, and what information goes into it. Include AI features switched on inside your existing software, and ask staff what they use on their own. Regulators now have a name for tools used outside a business's knowledge: "shadow AI". Their example is an employee using a tool they bought personally that doesn't meet the employer's policies.
2. Decide what stays out
One simple rule covers most of the risk: no personal or sensitive information goes into a tool you haven't approved for it. Canada's national cyber security agency gives the same advice, for security reasons.
| Type of information | Rule |
|---|---|
| Public or general material, such as a draft blog post or a generic email | Any tool the business allows |
| Your own internal, non-personal material | Approved tools only |
| Client or employee information | Only in a tool checked and approved for it, and only what's needed |
| Sensitive information, such as health, financial or identity details | Not without a privacy review first |
Often you can remove the name and details and still get what you need.
3. Choose the tool and account deliberately
Free personal accounts and business accounts can behave very differently. In their 2026 investigation of ChatGPT, regulators recorded that OpenAI's Enterprise, Team, Edu and API customer data is not used to train its models by default, while individual customers' interactions are. That's one vendor at one point in time, so check any tool you use, and look for an opt-out from training.
Our questions to ask before you buy an AI tool covers what to check.
4. Have a person review the output
AI can be wrong, and it can sound sure when it is. The Canadian Centre for Cyber Security warns that output "can be incorrect" and "can be biased". A person should review anything that goes to a client, gets published, or affects a decision about someone.
You also own what your AI says. In Moffatt v. Air Canada (2024), a BC tribunal held the airline responsible when its website chatbot gave a customer wrong information. It rejected the argument that the chatbot was responsible for its own actions.
5. Be open with people
Tell people when they're dealing with AI, for example a chatbot on your website. Regulators expect this, and they say public tools should warn people not to enter personal information unless it's necessary. Update your privacy policy if AI changes how you handle personal information.
6. Name one person, write one page, tell your staff
Name one person responsible for AI use, who can also pause a tool if something's wrong. Write a one-page rule covering which tools are allowed and what stays out. Then go through it with staff.
Alberta's commissioner calls the skill staff need "AI fluency": knowing the limits of the tools. Staff should also know that AI makes phishing messages more convincing, so an unusual request should be verified another way.
7. Check again when something changes
Look again when a tool adds a feature, a vendor changes its terms, or you start using a tool for something new. BC's commissioner warns that a transcription tool can quietly become a decision-support tool through an update. Regulators call this kind of drift "function creep". A yearly review of your list is enough for most small businesses.
Alberta and BC: consent is stricter
If you operate in Alberta or BC, this is the point most general guides miss.
In the ChatGPT investigation, the federal regulator accepted implied consent for one use under PIPEDA. The Alberta and BC commissioners said their Acts have no general "implied consent". BC has implicit consent and Alberta has deemed consent, and both have specific conditions. Either the purpose is obvious and the person gave the information voluntarily for it, or you give understandable notice and a reasonable chance to decline.
That finding was about OpenAI, not about small businesses. Our reading of how it carries over: if you want to put client information into an AI tool for a new purpose, such as analyzing client emails or training a custom assistant, don't assume consent. The purpose must be obvious, or you need proper notice with a chance to decline, or express consent.
Ontario: human rights also apply
Ontario's privacy commissioner and Human Rights Commission published joint AI principles in 2026. They're written for the public sector, and Ontario businesses generally fall under PIPEDA. But the Ontario Human Rights Code applies to private businesses, so an AI tool used to screen applicants or decide who gets a service needs watching for discrimination. Don't let a tool make those decisions alone.
When to do a privacy impact assessment
Regulators recommend a privacy impact assessment for AI that handles personal information, before you start and again when the tool or the vendor's terms change. It's usually not legally required for a private business, but it's the clearest way to show you thought it through.
You probably don't need one for a tool that only ever sees non-personal information. You probably do want one if a tool will touch client records, make recommendations about people, or send information outside Canada. Our guide on privacy impact assessments in Canada explains the process.
If something goes wrong
If someone puts client information into the wrong tool, or an AI tool exposes information, treat it as a possible privacy breach. Act the same day: stop the use, find out what went in and who could see it, and try to delete it.
Whether you must report depends on the facts and on which law applies:
- Federal law: assess whether there is a real risk of significant harm, and see PIPEDA breach reporting.
- Alberta: see Alberta breach notification.
- BC: BC PIPA has no breach reporting duty, but see what BC businesses should do.
What may be coming
None of this is law yet, and it may change:
- Alberta's commissioner has recommended a standalone AI law and rights around automated decisions.
- In September 2026, Canada's privacy regulators called on governments to legally require AI governance in health care.
Good habits now will put you ahead of any change. Privacy needn't hold you back either. As the federal regulator says, it "can promote innovation by building trust in new technologies."
Privacy case of the month
Once a month, a five-minute read: a real case a Canadian privacy commissioner investigated, what went wrong, and what to check in your own organization so it doesn't happen to you. Written for organizations under federal, Alberta and BC privacy law. Not written for Quebec.
We send one email to confirm. Nothing else arrives until you confirm, and an unconfirmed address is deleted after 7 days. Sent by ClearBreach Technologies Inc. Mailing address and contact details · Privacy policy
Sources
- Canada's federal, provincial and territorial privacy commissioners. Principles for responsible, trustworthy and privacy-protective generative AI technologies. December 2023.
- OPC, CAI, OIPC BC and OIPC Alberta. Joint investigation into OpenAI's ChatGPT. May 2026.
- OPC. AI, privacy, and your business.
- OIPC BC. PIPA and AI scribes. January 2026. Health-sector guidance, applied here by analogy.
- OIPC Alberta. Comments regarding responsible AI governance in Alberta. 2025.
- Canadian Centre for Cyber Security. Generative artificial intelligence, ITSAP.00.041. December 2025.
- BDC. New BDC study reveals 27% of Canadian entrepreneurs don't know they're using AI. September 2024.
- Ontario IPC and Ontario Human Rights Commission. Principles for the responsible use of artificial intelligence. January 2026.
- Federal, provincial and territorial privacy regulators. Responsible use of AI in health and health research. September 2026.
- Moffatt v. Air Canada, 2024 BCCRT 149. A small-claims tribunal decision, not binding on courts.
Related guides
- Questions to Ask Before You Buy an AI Tool. What to check with a vendor
- AI Note-Takers and Call Recording. The rules for tools that listen
- Privacy Impact Assessments in Canada. When and how to assess a new system
Frequently asked questions
Is it legal for a Canadian business to use tools like ChatGPT?
Yes. No Canadian privacy law bans a business from using AI. What the law controls is how personal information is handled. If you put a client's or employee's information into a tool, your privacy duties apply to that use, just as they do to any other system.
Can staff paste client information into a free AI tool?
They should not unless you have checked the tool and approved that use. Canada's privacy regulators say personal information should go into a prompt only where that is authorized. Set a simple rule, and give staff an approved tool for work that needs one.
Do I need a privacy impact assessment before using AI?
Alberta PIPA, BC PIPA and PIPEDA do not generally require a private business to do one. Canada's privacy regulators still recommend it for AI that handles personal information, and it is a sensible step before a tool touches client records. For a tool that handles no personal information, it is usually unnecessary.
Does a business account make an AI tool compliant?
No. A business account can give you better settings, and some vendors do not train their models on business customers' data by default. But that varies by vendor, and your duties around purpose, consent, accuracy and security still apply.
This guide is educational and does not constitute legal advice. It is grounded in the text of PIPEDA, Alberta PIPA, and BC PIPA and published guidance from the OPC, OIPC Alberta, and OIPC BC. If your situation involves regulatory investigation, litigation risk, or circumstances not addressed here, engage a qualified privacy lawyer.
Check an AI tool before client data goes in
See a completed privacy impact assessment for a new system: the risks identified, the mitigations recommended, and the record it leaves behind.
See a completed PIA →