Key takeaways
- The Cyber Centre's generative AI guidance, ITSAP.00.041, was republished in December 2025 and lists AI usage policies among its mitigations.
- Keep personal information, account numbers, passwords and one-time codes out of prompts.
- Approve tools deliberately, and vet vendors before staff use them.
- Require a person to review any AI output that affects customers or money.
- Build on the Cyber Centre's 13 baseline controls. CyberSecure Canada certification is voluntary.
- Name an owner for the policy
- List approved AI tools and how to request a new one
- Ban customer personal information, account numbers, passwords and codes from prompts
- Vet each vendor before approving its tool
- Require human review of outputs that affect customers or money
- Turn on multi-factor authentication and train staff before giving access
- Set a date to review the policy
Why does your business need an AI policy now?
The Canadian Centre for Cyber Security (the Cyber Centre) republished its generative AI guidance, ITSAP.00.041, in December 2025, replacing the July 2023 version. AI usage policies are one of the mitigations it recommends.
A policy doesn't need to be long. Its job is to answer three questions for everyone in the business: Which tools can I use? What can I put into them? Who checks the output before it matters?
A written policy also helps when something goes wrong. If someone pastes sensitive information into a tool, a clear rule and a set incident step let you respond quickly instead of working it out on the spot. And when a customer, supplier or your bank asks how your business uses AI, you'll have a ready answer.
What risks does the Cyber Centre name?
ITSAP.00.041 lists these risks. In plain terms:
- Misinformation: answers that sound confident but are wrong.
- Phishing: more convincing scam messages written with AI.
- Privacy leakage: personal or confidential information typed into prompts.
- Malicious or buggy code: generated code that doesn't do what it seems to, or does harm.
- Poisoned datasets: training data that has been tampered with to skew results.
- Bias: outputs that treat people unfairly.
- Loss of intellectual property: your own content or ideas exposed through a tool.
The Cyber Centre's mitigations
Multi-factor authentication, patching, staff training, AI usage policies, vetting vendors, and keeping personal information out of prompts.
Which privacy principles apply to generative AI?
On December 7, 2023, federal, provincial and territorial privacy regulators issued joint principles for responsible, trustworthy and privacy-protective generative AI. Their headings, listed below, make a handy checklist for a small business.
Several translate directly into everyday rules. Have a proper basis, including consent where needed, before putting customer information into a tool. Use only the information the task needs. Be open with customers when AI plays a meaningful part in how you deal with them. Check AI output for accuracy before relying on it. And make sure one person answers for how tools are used. Our PIPEDA guide explains the federal privacy principles in more detail.
- Legal authority and consent
- Appropriate purposes
- Necessity and proportionality
- Openness
- Accountability
- Individual access
- Limiting collection, use and disclosure
- Accuracy
- Safeguards

What goes in a one-page AI use policy?
Here's an outline you can adapt. Keep it to one page so people actually read it, and write it in your own words.
- Purpose and owner: why the policy exists and who is responsible for it.
- Approved tools: the AI tools staff may use for work, and how to ask for a new one.
- Data that never goes into prompts: customer and staff personal information, bank account and card numbers, passwords and one-time codes, payroll and tax details, and confidential contracts.
- Vendor vetting: questions to answer before approving a tool, such as where data is stored, whether prompts are kept or used to train the tool, and how accounts are secured.
- Human review: any AI output that affects customers or money, such as quotes, invoices, payment instructions, customer messages or financial figures, is checked by a person before it is used.
- Security basics: multi-factor authentication on AI accounts, business accounts rather than personal ones, and training before access.
- Incidents: what to do if something sensitive goes into a tool or an output causes a problem.
- Review date: when the policy will be looked at again.
How do you put the policy to work?
A policy is easier to follow when people can see how it applies to real tasks. Consider adding a few short examples like these, adjusted to your business. They're illustrations, not rules from any regulator.
| Task | OK? | Condition |
|---|---|---|
| Drafting a general marketing email | Yes | Use an approved tool and have a person review it before it goes out. |
| Summarizing a customer complaint | Only with care | Remove names, contact details and account numbers first. |
| Checking figures in an invoice batch | As a draft only | A person checks every total against the source documents. |
| Pasting a customer list to segment it | No | Customer personal information stays out of prompts. |
| Writing payment instructions for the bank | No | Payment details are prepared and approved through your usual process. |
What extra rules apply to banking and payments?
AI tools and payments call for a few rules of their own:
- Never paste online banking credentials, one-time codes or full account numbers into an AI tool.
- Don't let AI-drafted text change payment instructions without your usual verification and approval.
- Treat AI summaries of bank statements or invoices as drafts, and check the figures against the source.
- Check your banking agreement before connecting any outside app to your accounts.
- Remember that AI also helps fraudsters write convincing messages. Our small business fraud guide covers verification habits.
How do banks prepare their own staff?
Large banks have described how they handle this internally. CIBC says employees must complete a mandatory course on its AI approach and governance before using its in-house platform, CIBC AI. BMO describes an AI for All program that teaches every employee practical AI skills, including responsible use and governance.
You don't need a formal program, but the idea scales down well: explain the policy and the reasons behind it before you give anyone access to a tool. Refresh that conversation whenever you add a tool or change the rules.
Which baseline controls should back up your policy?
A policy is far more effective on top of solid basic security. The Cyber Centre's baseline cyber security controls for small and medium organizations (fewer than 500 employees) set out 13 controls, including strong authentication, employee training, access control, secure cloud and outsourced IT, and an incident response plan. Many generative AI tools are cloud services, so the cloud and access-control items apply directly.
If you want outside recognition, CyberSecure Canada is a voluntary certification program for small and medium-sized organizations. ISED stopped being the program authority on March 31, 2023, and current information is with the Standards Council of Canada.
Not sure where to begin? Our small business guide to AI in banking covers your bank's side, and the business banking page lists the AI tools banks offer business clients.
What the Canadian Bank AI Index shows on this topic, counted from the published data. Every number can be rechecked in the database.
| Institutions scored | 35Canadian Bank AI Index, 2026 Q4 |
|---|---|
| Entries in the database | 210Sourced and dated, as of 2026-11-01 |
| Entries filed under Business banking | 7 |
| Of those, reaching customers | 1Customer-facing or running on customer accounts |
| Live (launched or disclosed in use) | 1Pilots and announcements excluded |
| Institutions with a live, sourced customer use | 1 of 35 (3%)TD |
Open these entries in the database →Computed November 1, 2026.
What this means for you
- Start by asking your team which AI tools they already use and for what, then list the approved tools and how to request a new one.
- Before approving a tool, find out where data is stored, whether prompts are kept or used to train it, and how accounts are secured.
- Use business accounts with multi-factor authentication for AI tools, and walk staff through the policy before giving them access.
- Never let AI-drafted text change payment instructions without your usual verification, and treat AI summaries of statements or invoices as drafts to check against the source.
- Back the policy with the Cyber Centre’s baseline controls, and see our small business fraud guide for verification habits.
How to write a one-page generative AI policy for a small business
- 1Find out current useAsk your team which AI tools they already use and for what tasks.
- 2Choose approved toolsVet each vendor, including where data is stored and how accounts are secured, then list the tools staff may use.
- 3Define off-limits dataList what never goes into prompts: personal information, account and card numbers, passwords, one-time codes and confidential documents.
- 4Set review rulesRequire a person to check any AI output that affects customers or money before it is used.
- 5Secure the accountsTurn on multi-factor authentication and use business accounts for AI tools rather than personal ones.
- 6Train and shareWalk staff through the policy before giving access, and keep it where everyone can find it.
- 7Review on scheduleSet a review date and revisit the policy when you add a tool or have an incident.
Questions people ask
Does a small business need a long AI policy?
No. A one-page policy covering approved tools, off-limits data, vendor checks and human review addresses the main points in the Cyber Centre's guidance.
What should never go into an AI prompt?
Personal information about customers or staff, bank account and card numbers, passwords, one-time codes and confidential documents. The Cyber Centre's guidance specifically advises against putting personal information in prompts.
Is CyberSecure Canada certification mandatory?
No. It's a voluntary certification for small and medium-sized organizations, and current program information is with the Standards Council of Canada.
Do privacy obligations still apply when staff use AI tools?
If staff put personal information into an AI tool, your business's privacy obligations still apply to how that information is handled. Canada's privacy regulators issued joint generative AI principles in December 2023 that cover consent, purposes, safeguards and more.
How often should we review the policy?
Set a review date when you write it, and revisit the policy whenever you add a new tool or something goes wrong.
Sources
Official and first-party pages, with the date shown on each page.
- 01Canadian Centre for Cyber Security — Generative artificial intelligence (AI) - ITSAP.00.041Page date: December 2025
- 02Canadian Centre for Cyber Security — Baseline cyber security controls for small and medium organizationsPage date: February 18, 2020
- 03Office of the Privacy Commissioner of Canada — Principles for responsible, trustworthy and privacy-protective generative AI technologiesPage date: December 7, 2023
- 04Innovation, Science and Economic Development Canada — CyberSecure CanadaPage date: September 24, 2025
- 05CIBC — CIBC launches in-house AI platform enterprise-wide after successful pilotPage date: May 27, 2025
- 06BMO — BMO Earns Top Recognition for AI Talent Development in Global BankingPage date: October 23, 2025
Educational information only, not financial, legal or privacy advice. Rules and bank features change; check the linked official pages and talk to your financial institution or a qualified professional about your situation.
