AI Regulation Tracker  /  Professional guidance

CPA Ontario: Accountants Must Stand Behind Every AI Output in Audit and Assurance Work

CPA Ontario's regulatory guidance says artificial intelligence does not shift professional accountability. A Chartered Professional Accountant must still be able to explain, support and defend any AI-generated figure used in audit, assurance or reporting.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

CPA Ontario has told the province's Chartered Professional Accountants that artificial intelligence changes the tools of the job but not the responsibility. In regulatory guidance titled "Accountabilities for CPAs in the Age of Artificial Intelligence," current for 2026, the regulator sets a plain expectation: a CPA who uses AI in audit, assurance or reporting must be able to explain, support and stand behind the output. The professional accountability for the work, and for the opinion, stays with the accountant.

The message matters because AI is already inside accounting workflows. Tools now draft reconciliations, flag anomalies, summarize contracts and propose journal entries, often faster than a junior associate could. CPA Ontario's position is that none of this dilutes the member's duty. AI does not remove the obligation to test transactions, assess accuracy and own the audit opinion. The tool can help find the answer, but the accountant is answerable for it.

The guidance also reflects a known weakness in current AI systems. Generative models can produce output that looks precise and authoritative while being wrong, and they do not reliably show their reasoning. In an audit context, a confident but incorrect figure is a specific risk: it can pass a quick review precisely because it looks finished. CPA Ontario's framing puts the burden on the member to catch that, not on the tool to prevent it.

The accountability does not transfer

The core of the guidance is a line about where responsibility sits. When a CPA relies on an AI output in assurance work, that output is not a finished conclusion to be passed through. It is an input the accountant must be able to justify. If a regulator, a client or a court asks how a figure was reached, "the model produced it" is not an answer that satisfies professional obligations. The member must be able to explain the basis for the output and support it with evidence gathered through their own procedures.

What it still requires

The guidance reinforces duties that predate AI rather than inventing new ones. Testing transactions, evaluating the accuracy and completeness of information, exercising professional skepticism and forming an independent opinion remain the accountant's work. The practical effect is that AI use has to be paired with verification. An accountant who accepts an AI-generated schedule without checking it against source records has not met the standard simply because the technology is sophisticated. CPA Canada and the AICPA make a parallel point in their assurance material on closing the AI trust gap, which treats explainability and human oversight as conditions for trusting AI-assisted work.

What it does not do

The guidance does not ban AI in accounting, and it does not tell members which tools to buy. It is not a prescriptive technical standard with a checklist of approved systems. It also does not create a new category of penalty on its own. It is regulatory guidance that clarifies how existing professional-conduct expectations apply when AI is in the workflow. That is its reach and its limit: it tells accountants that the familiar duties still bind, and that AI is judged against them.

The cross-border read

For a US accountant, the direction is instructive rather than binding. It parallels the emphasis in the PCAOB's QC1000 quality-control standard and AICPA expectations that human professionals remain accountable for audit quality regardless of the technology used. An Ontario member is governed by CPA Ontario, but any accountant relying on AI in assurance work faces the same underlying principle from their own regulator: the opinion is yours, so the verification has to be yours too. The specifics differ by jurisdiction; the standard of ownership does not.

The takeaway for practitioners is operational. Use AI where it helps, but build a record that shows you tested the output, understood how it was produced and reached your own conclusion. In practice that means documenting which tool was used, what it produced, how the result was checked against source records, and why it was fit to rely on. The guidance treats that discipline as the price of relying on AI in work that carries a professional signature, and it makes clear that the signature, not the software, is where responsibility comes to rest.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.