Written by Mike Pearlstein, CISSP, CEO of Fusion Computing Limited. Helping Canadian businesses build and manage secure IT infrastructure since 2012 across Toronto, Hamilton, and Metro Vancouver.
This is the clause-by-clause AI policy playbook Canadian accounting firms can adapt and adopt under the CPA Code of Professional Conduct. Every clause maps to a named CPA Ontario expectation from Accountabilities for CPAs in the Age of Artificial Intelligence (2024), to a relevant Canadian Standard on Quality Management 1 (CSQM 1) quality objective or firm-designed response, and to the privacy regime the firm operates under.
Those regimes are PIPEDA federally, Quebec Law 25, and the provincial private-sector privacy acts in Alberta and British Columbia. The download is a fill-in AI Acceptable Use Policy template, the CPA-safe AI deployment guide, and the full fourteen-clause Word edition. All three arrive by email.
Why the CPA Code drives every clause
According to CPA Ontario’s Code of Professional Conduct (current consolidation), Rule 208.3(a) requires a member or firm to “take appropriate measures to maintain and protect confidential information” and to limit access to those with legitimate purpose. Rule 208.3(b) then requires written agreement from anyone granted that access. Every clause in this template starts from those two sentences.
The template is anchored to CPA Ontario’s Accountabilities for CPAs in the Age of Artificial Intelligence (2024) and to CPA Canada’s CSQM 1. CSQM 1 applied from 15 December 2022 for audits, reviews and other assurance engagements, and from 15 December 2023 for related services engagements. It also maps to the federal and provincial privacy statutes that govern client data flowing into a model. It was also reviewed against CPA Ontario’s June 2026 guidance, The Responsible Use of AI in Professional Practice, which confirms that existing CPA Code duties continue to apply to AI-assisted work and that AI-generated output must be reviewed and verified before use.
In the Word edition, every clause cites its source obligation, so a CPA Ontario practice inspector can trace each control back to a named authority.
The policy maps to four CPA Code rules. Rule 202 (integrity and due care) applies because the AI output is the work the partner signs. Rule 203 (professional competence) requires the firm to demonstrate it understands what the tool does.
Rule 205 (false or misleading documents) is where AI hallucination becomes a discipline risk. Rule 208 (confidentiality of information) is the constraint on what client data can flow into a model at all.
The approved-tools matrix names each product, from Microsoft 365 Copilot down to consumer chatbots, against the engagement type it may touch. A managing partner can hand it to IT as the firm’s working starting point.
What the policy template covers
The matrix covers Microsoft 365 Copilot inside the firm tenant (the FC-recommended default), CCH iFirm AI assistants, Caseware AiDA and IDEA for audit, and where personal and consumer AI accounts (consumer ChatGPT, Claude.ai, and Gemini) are prohibited for client data unless the exact plan, settings, and contract terms meet the firm’s approved-tool standard.
The 90-day rollout plan starts with tenant readiness: Microsoft Purview sensitivity labels, Conditional Access sign-in rules, and Microsoft Entra ID identity controls, which Fusion maps to CRA’s current EFILE confidentiality and security-best-practice obligations. CSQM monitoring design and the four-hour competence session follow.
The Word edition adds a monitoring-record template so the firm can evidence its CSQM 1 monitoring.
The work-product-verification clause addresses a documented failure mode in Canadian court records: fabricated citations reaching filed work. In tax work, the analogous risk is an unsupported CRA folio or calculation reaching a client file or return. When a model fabricates a tax citation, a financial-statement footnote reference, or an audit working-paper conclusion, the partner who signed the file owns the result under Rule 205.
How firms apply this to their existing AI work
The Word edition’s clause requires the preparer to sign a per-engagement attestation confirming every AI-assisted citation or calculation in a working paper was independently verified against the primary source (a CRA folio, the CPA Canada Handbook or the underlying client record) before the partner approves the file.
Without that attestation, the firm loses one clear record that the AI-assisted citation or calculation was independently checked. The firm’s CSQM 1 monitoring process can still inspect the underlying engagement file and other evidence. I would rather a Canadian firm adopt a short policy it actually follows than a long one my own team drafted for it.
Cyber-renewal questionnaires can include governance questions about AI and information security. Check the current questionnaire with your broker; premium outcomes depend on the firm’s full risk profile.
What is a CPA AI policy, and how can it document existing CPA Code duties?
According to the Canadian Centre for Cyber Security (2023), prompts can move organization-controlled information to a service provider, and unauthorized use of generative AI can bypass approved data-governance frameworks. Fusion’s recommendation is to define approved tools, approved uses, and approved data classes before deployment. A CPA AI policy is that definition, written in CPA Code language so an inspector can follow it.
A CPA AI policy is a short governing document, usually 8 to 14 clauses, that answers four questions in writing. Which tools may touch client data. Which engagement classes may use them. Who supervises the output. And how the firm evidences all three through its own CSQM 1 monitoring process.
In my reading of the 2024 paper, CPA Ontario has issued no rule that says “write an AI policy.” What it has said, through the 2024 accountabilities paper, is that the existing rules already reach AI-assisted work. Rule 208 governs the data going in. Rule 205 governs the document coming out. A written policy is one practical way to document how the firm applies both.
In our practice the clause that gets cut first is the verification attestation, and it is a clause a practice inspector can examine. Our engineers found that the binding constraint on a Copilot rollout is rarely the model. It is the SharePoint permission set the firm inherited from an old file-server migration.
The tenant-readiness work has to happen before any AI clause is enforceable: reviewing Microsoft Purview sensitivity labels and Conditional Access scales with firm size, existing permissions, and the tenant remediation the review turns up. The policy text itself is the smaller job.
Who is this for?
According to Microsoft’s Copilot data and privacy documentation (2026), Microsoft 365 Copilot returns only content the signed-in user already has permission to open, so existing permission sprawl becomes answer sprawl. That is why the approved-tools clause and the tenant-permissions work belong in the same policy.
This template is for the managing partner of a 4 to 50 staff Canadian accounting firm who is being asked whether the firm has a written AI policy. That question comes from an audit client’s in-house controller during procurement, a cyber insurance broker at renewal, the person performing the firm’s CSQM 1 monitoring, or a senior associate who just read CPA Ontario’s 2024 paper.
A one-line rule like “don’t paste client data into public chatbots” is a start. It doesn’t name the approved tools, the reviewer, or the evidence the firm keeps. This template does. Start with the policy, then fix the Microsoft 365 permissions that would make the policy hard to enforce.
It is also for the CFO or IT director at a 50 to 200 staff regional firm who needs a reviewable starting point rather than a blank Word document.
I wrote the clauses so a solo practitioner can strike out what does not apply without breaking the CSQM 1 logic. Solo practices can tailor monitoring to their size, but completed-engagement inspections still have to meet CSQM 1 objectivity requirements. A practitioner who worked on the engagement, or served as its engagement quality reviewer, cannot inspect that same engagement. A sole practitioner may need an eligible external monitor. Engagement quality review is a separate CSQM 2 process where applicable.
It is not intended for non-Canadian firms without adaptation. US firms operating under AICPA standards should cross-check the AICPA AI guidance and their state board rules. Cross-border firms should adopt the strictest applicable standard rather than the CPA Ontario baseline, and the Word edition’s footnotes name the divergence points.
Quebec-only firms should pay particular attention to the Law 25 transfer clause, which requires a documented assessment before client personal information leaves the province for a model hosted elsewhere.
CPA AI policy template vs a generic AI policy: the difference the CPA Code requires
According to Zhang v. Chen, 2024 BCSC 285 (2024), Justice Masuhara dismissed a request for special costs and held counsel personally liable for costs under Supreme Court Family Rule 16-1(30) after two ChatGPT-invented cases reached a court filing. The lesson maps onto CPA Code Rule 205: the person who signs owns the output.
A generic AI policy tells staff to be careful and to check their work. It reads well and evidences nothing. A CPA-specific policy names the rule (202, 203, 205, 208), the tool, the engagement class, the supervising partner and the artifact that proves each one at review time.
When I review a draft policy for a Canadian firm, my first pass is a single question asked of every clause: what document does this produce, and who files it? Clauses that produce no artifact get cut, because the person performing the firm’s CSQM 1 monitoring cannot examine an intention.
The same court decision cites a January 2024 study, Large Legal Fictions (2024), which measured legal hallucination rates between 69% and 88% across the models it tested. Those are legal citations, and the equivalent failure in a tax file is a fabricated CRA folio reference in a working paper a partner has already signed.
If you want help with the Microsoft 365 Copilot configuration that sits behind the template’s data-classification rules, the Microsoft Purview sensitivity labels that we recommend putting in place before a broad Copilot rollout, or the Entra ID hardening that Fusion maps to CRA’s EFILE confidentiality and security-best-practice obligations, that is work Fusion Computing does for Canadian accounting firms every week. Book a CPA AI policy working session with the partner who would own the engagement.
Download the CPA AI Policy Template
Fill in the three fields below. We will email the fill-in AI Acceptable Use Policy template as a PDF plus the CPA-safe AI deployment guide, with a direct download link on screen as soon as you submit. An editable Word edition ships with every download.
FREE DOWNLOAD
The CPA AI Policy Template for Canadian Accounting Firms (2026)
Written by Mike Pearlstein, CISSP. No sales call required.
Form not loading? Email us directly and we’ll send the template the same business day.
According to the Personal Information Protection and Electronic Documents Act (2000), section 10.1 requires a breach report to the Privacy Commissioner as soon as feasible, with no fixed 72-hour clock in the statute. The template privacy clause names who decides, who reports, and where the record lives.
Related deep dives
According to Quebec’s private-sector privacy act (2023), a firm must assess the privacy implications before communicating personal information outside the province. Cross-province accounting firms need that assessment on file before any model hosted outside Quebec touches client records, and the sibling posts below cover the configuration work that assessment usually turns up.
- The full AI for Canadian Accounting Firms walkthrough: the CPA Ontario rules, the CSQM 1 framework, and the clause-by-clause logic behind every section of the template.
- Microsoft 365 Copilot vs Generic ChatGPT for Canadian CPA firms: the approved-tools decision matrix expanded into a head-to-head review of why tenant-scoped Copilot is the default and consumer ChatGPT is the prohibited tool.
- CRA EFILE security for Canadian accounting firms: how the template’s data-classification rules translate into actual Microsoft 365 configuration, Entra ID hardening, and the access controls Fusion maps to CRA’s EFILE confidentiality, credential-protection, and security-best-practice obligations.
- CCH iFirm and CaseWare cybersecurity hardening for Canadian accounting firms: the engagement-system piece of the approved-tools matrix and the data-segregation pattern that keeps AI exposure inside the firm tenant.
- Tax-season cybersecurity for Canadian CPA firms: the operational pressure context that explains why a written AI policy filed before February matters more than one drafted in May.
- FINTRAC IT controls for Canadian accountants: the reporting-entity obligations that sit alongside the CPA Code when a firm is also a FINTRAC reporting entity.
Frequently Asked Questions
According to the Office of the Privacy Commissioner of Canada (2018), an organization must report a breach creating a real risk of significant harm and keep a record of every breach for 24 months. These are the questions Canadian firms ask us before they download.
What’s the download?
A fill-in AI Acceptable Use Policy template as a PDF, plus the CPA-safe AI deployment guide for Canadian accounting firms. An editable Word edition is included, so you can adapt the clauses directly rather than retyping them.
Does the template cover Microsoft 365 Copilot specifically, or only consumer tools?
Both. The fill-in template ships with an approved-tools table, with tenant-scoped Microsoft 365 Copilot as the worked example, because it answers only from files the signed-in user can already open. The prohibited-tools list rules out free and personal-tier consumer AI accounts for client data, which covers consumer ChatGPT, Claude.ai, and Gemini. The fourteen-clause Word edition maps twelve engagement scenarios, from T1 preparation through to audit risk assessment.
How long does adoption take, and who has to sign?
The recommended rollout runs 5 steps across 90 days: tenant readiness, CSQM 1 monitoring design, a 4-hour competence session, written sign-off filed with personnel records, and intranet publication. Every partner and every preparer signs once at adoption and again at the annual re-attestation Fusion recommends. Timing depends on firm size, approvals, existing permissions, and tenant remediation; a firm that only needs the policy filed can scope it separately from the Microsoft 365 readiness work.
How will my data be used?
Your name, firm name, and email go into Fusion Computing’s contact system, and we email the template files after submission. We may send updates on Canadian accounting firm IT, CPA Ontario AI guidance, and CSQM 1 developments, no more than once a month.
We do not sell your contact information. We use service providers, including our CRM and email systems, to process the form and deliver the files. Any ongoing marketing email includes an unsubscribe link. See our Privacy Policy for details.
Is this just a sales pitch?
No. The download is the deliverable, and it works without our involvement. You do not need to speak to us to use it.
We make the policy free because regulator-anchored documents like this one are how Canadian accounting firms find out Fusion Computing exists. If you later want help with the Microsoft 365 Copilot rollout, the Purview labels, or the CSQM 1 monitoring system, you can reach out on your own timeline.
Do I need to be an existing FC client?
No. The template is free for any Canadian accounting firm, in-house finance team, or accounting academic to download and adapt.
Use it and modify it freely inside your firm; we ask that you keep the Fusion Computing attribution and not resell it.
Can I share it with my partner or colleague?
Yes. Share it with anyone inside your firm, with outside CPA counsel, with the person who performs your CSQM 1 monitoring, with your cyber insurance broker, or with your IT vendor. We ask that the attribution to Mike Pearlstein and Fusion Computing stay on the title page. Beyond that, the template is designed to be marked up, struck through and rewritten.
Who wrote this?
Mike Pearlstein, CISSP, CEO of Fusion Computing Limited. Fusion Computing has been doing regulator-anchored AI deployment work for Canadian accounting firms, law firms, financial brokerages, and healthcare clinics since 2012.
The template was reviewed against CPA Ontario’s 2024 accountabilities paper and its June 2026 guidance (The Responsible Use of AI in Professional Practice), CPA Canada’s CSQM 1 guidance, the CRA EFILE 2025 security expectations, FINTRAC’s reporting entity obligations, PIPEDA, and the provincial privacy acts in Alberta, British Columbia, and Quebec.
Bottom line
CPA Ontario has not amended the CPA Code to mandate a written AI policy. Rules 202, 203, 205, and 208 already reach every engagement an AI tool touches. A written policy, a filed monitoring record, and an annual partner re-attestation are the operating model Fusion recommends for showing it; no standard prescribes those exact steps.
