How the CRA uses artificial intelligence in Canadian tax audits: What taxpayers must know now

CRA Artificial Intelligence and Tax Audit Risk: Overview for Canadian Taxpayers

David J Rotfleisch, CPA, JD is the founding tax lawyer of Taxpage.com and Rotfleisch & Samulovitch P.C., a Toronto-based boutique tax law corporate law firm.

Artificial intelligence is no longer a future consideration for Canadian taxpayers. It’s already embedded in how the Canada Revenue Agency selects returns for tax audit, enforces compliance, and administers penalties. The CRA actively deploys machine learning algorithms, predictive data analytics, and AI-driven cross-referencing tools to identify anomalies in filed tax returns, flag high-risk taxpayers, and detect unreported income from cryptocurrency transactions, offshore structures, platform-economy earnings, and digital payments.

Every Canadian who files a tax return is, in some measure, already being assessed by an automated system before a human CRA officer ever opens the file.

The implications are significant. A taxpayer whose return is algorithmically flagged as an outlier faces a materially elevated risk of tax audit, tax reassessment, and penalties — often without any knowledge that an automated system drove the process. Meanwhile, the CRA’s parallel use of AI in taxpayer services has produced a cautionary lesson about the gap between government claims and independent verification. The CRA’s chatbot Charlie cost taxpayers $18 million to build and operate over five years.

When the Auditor General independently tested Charlie in October 2025, it was still the original rule-based scripted tool — not yet a true generative AI system — and its performance fell well short of even the CRA’s own internal benchmarks. The CRA subsequently upgraded Charlie to a generative AI version in November 2025, claiming approximately 90% accuracy in pre-release testing, but that figure remains independently unverified. The disparity between internal claims and independent findings is itself a reason to scrutinize CRA AI enforcement tools with equal rigor.

For Canadian taxpayers, business owners, self-employed professionals, and cryptocurrency investors, understanding where CRA artificial intelligence operates, what it targets, and what legal protections remain available is now a foundational element of responsible tax planning. Experienced Canadian tax lawyers are already seeing the real-world impact of AI-driven tax enforcement in the files they handle — and the time to understand this landscape is before a CRA tax audit letter arrives, not after.

CRA Tax Enforcement and AI Technology: Background and Historical Context

The CRA’s adoption of artificial intelligence and advanced data analytics has been underway for several years, but practical deployment in compliance and enforcement has accelerated sharply. The CRA has incorporated digital solutions and enhancements across its entire continuum of activities — from tax reporting to compliance interventions — with the aim of improving the tax system’s efficiency and effectiveness. International engagement shows the CRA is not unique in experiencing these pressures, and the agency has been exploring more fundamental ways to modernize and automate its tax processes.

On the taxpayer service side, the CRA has focused on using robotic process automation to automate high-volume repetitive tasks in the processing of tax credits and rebates, allowing employees to focus on higher-value work while improving speed and accuracy. Its chatbot “Charlie” — first launched as a rule-based scripted tool in February 2020, initially to answer COVID-19 benefit questions — is the most visible example of this approach. Although the CRA promoted Charlie as a modern digital solution to help taxpayers navigate the tax system and reduce demand on overburdened call centres, the chatbot could only answer general tax questions from pre-determined scripts and was still described as “learning about the CRA” at launch.

The chatbot’s real-world accuracy has been a matter of significant dispute, and the timeline matters. The CRA has long defended Charlie on the basis that internal tests showed it had a 70% accuracy rate — meaning nearly one in three responses were still incorrect. In October 2025, the Auditor General tested the then-current version of Charlie — still the original rule-based scripted tool that answered from pre-determined scripts rather than generating new content — and found its performance well short of even that internal benchmark. The Auditor General found that Charlie offered correct information only about one-third of the time. Following the audit, in November 2025 the CRA upgraded Charlie to a generative AI version similar to tools like ChatGPT, claiming pre-release testing showed an accuracy rate of approximately 90%, but acknowledged it cannot confirm real-world accuracy without reviewing every interaction. The generative AI version’s true real-world performance therefore remains independently unverified.

Over five years, the CRA paid $18 million for Charlie — a chatbot the Auditor General says gave her team the wrong answer approximately two-thirds of the time. The divergence between the CRA’s internal accuracy claims and the Auditor General’s independent findings is instructive for a broader reason. CRA Assistant Commissioner Melanie Serjak told MPs that the agency plans to integrate AI-based tools to help agents deliver more reliable tax information to Canadians, alongside a revamped training model. Whether those improvements close the gap between claimed and verified performance remains to be seen — and that uncertainty is directly relevant to how Canadian taxpayers should assess the reliability of AI tools the CRA deploys on the enforcement side as well.

Despite these service-side difficulties, the CRA has simultaneously invested heavily in AI tools aimed squarely at compliance enforcement — where the stakes for Canadian taxpayers are considerably higher than a misanswered general tax question. The CRA is relying on machine learning and AI to sift through large amounts of data to help it decide whose tax returns to audit. This dual track — AI for service and AI for enforcement — means that Canadians now interact with artificial intelligence at virtually every stage of their relationship with the CRA, from the moment they file a return to the moment they may receive a tax audit notice.

How CRA AI Selects Canadian Tax Returns for Audit: Key Issues and Findings

AI-Powered Tax Audit Selection and Predictive Risk Modelling

The CRA’s integration of AI and digital tools is part of a strategic effort to modernize its compliance framework. Key components include AI-powered data analytics, where CRA algorithms cross-reference tax returns with third-party data, identifying anomalies or patterns that may indicate underreporting or aggressive tax planning. Using data mining and predictive modelling, the CRA prioritizes tax audits and enforcement actions based on risk levels, allowing a more efficient allocation of resources toward potential non-compliance. Complex tax arrangements, including offshore structures and cryptocurrency transactions, can now be detected more effectively through AI-driven analysis.

The CRA has stated that it will continue to improve the use of data, technology, and business intelligence to more accurately and efficiently identify high-risk taxpayers. Using data analytics and advanced risk assessment, the CRA aims to better identify and address high-risk aggressive tax planning.

This means that AI is playing a decisive — if opaque — role in the selection of taxpayers for tax audit. A taxpayer whose return is flagged by an algorithm as anomalous may have no direct awareness that an automated system was involved in triggering their tax audit, and no straightforward legal avenue to discover it.

CRA’s Official Position: Human Oversight of AI-Assisted Tax Decisions

Despite these developments, the CRA has been careful to maintain that AI does not make final decisions about individual Canadians. A CRA spokesperson has stated that all tax filings are reviewed and processed with human oversight, and that AI does not determine audits, assessments, or eligibility for benefits. Instead, the agency says it is exploring the responsible use of AI.

Government employees are prohibited from entering personal taxpayer data into publicly available AI tools, a restriction intended to prevent sensitive information from being exposed to external systems that may not meet Canadian privacy standards. Before any federal department can adopt or develop an AI system, it must go through formal privacy and risk assessments, including consulting privacy officials and ensuring compliance with national standards.

This distinction between “AI-assisted” and “AI-determined” decision-making is legally significant. Where AI recommends and a human confirms, the traditional framework of administrative law — including the right to a fair hearing and the right to know the basis of a decision — may still apply in principle. But where AI systems operate in ways that are opaque even to the human officers nominally supervising them, those protections risk becoming theoretical rather than real.

The Federal Regulatory Framework: Directive on Automated Decision-Making

The federal government has attempted to address these concerns through its Directive on Automated Decision-Making (the “Directive”), administered by the Treasury Board Secretariat. The Government of Canada has committed to using artificial intelligence in a manner that is compatible with core principles of administrative law such as transparency, accountability, legality, and procedural fairness.

The Directive requires departments to ensure that testing and monitoring assesses human rights and is consistent with applicable legislation such as the Canadian Charter of Rights and Freedoms, the Canadian Human Rights Act, and the United Nations Declaration on the Rights of Indigenous Peoples Act. It also requires documenting client feedback, unexpected impacts, human overrides of decisions made by the system, and other system failures.

Critically, under the Directive, departments must complete an algorithmic impact assessment before deploying automated tools. Existing automated decision systems developed or procured prior to June 24, 2025, will have until June 24, 2026 to comply with the new or updated requirements. This means many CRA AI systems are currently in a compliance transition period — a gap that experienced Toronto tax lawyers and administrative law practitioners are watching closely.

CRA AI and Cryptocurrency: Heightened Tax Audit Risk for Digital Asset Holders

The CRA’s increasing reliance on AI and digital analytics has several implications for taxpayers. E-transfers, cryptocurrency holdings, and online payments are now routinely analyzed. This is particularly significant for Canadian cryptocurrency holders and investors, whose reported income is now cross-referenced by CRA AI against data collected from financial institutions, exchanges, and third-party payment processors. Unreported or underreported cryptocurrency gains are among the highest-risk items in CRA’s current AI-driven enforcement environment.

AI Accuracy Concerns and Practical Risks for Taxpayers

The Charlie experience raises a pointed practical question about enforcement-side AI. Where the CRA’s own internal benchmarks consistently diverged from what the Auditor General independently verified on the service side, taxpayers are entitled to ask what meaningful assurances exist that the AI tools selecting their returns for tax audit are performing as claimed. An erroneously flagged tax audit is not a theoretical inconvenience — it means professional fees, management time, production of years of records, potential tax reassessments, and the stress and disruption of a process that can take months or years to resolve. A taxpayer subjected to a tax audit triggered by an inaccurate or biased algorithm, and who cannot discover or challenge the algorithmic basis for that selection, has no practical remedy at the front end of the process. The only recourse available is to mount a full defence after the fact — at considerable cost.

AI-driven automation, predictive analytics, and natural language processing enhance administrative efficiency and revenue collection but raise concerns regarding data security, legal accountability, and algorithmic bias. Policies balancing innovation with robust governance mechanisms are essential to harness AI’s benefits while mitigating risks.

CRA AI Tax Enforcement: Implications for Canadian Taxpayers and Businesses

The CRA’s deployment of AI for tax audit selection and compliance enforcement has several practical implications that all Canadian taxpayers and their advisors should understand:

• Tax audit triggers are increasingly algorithmic. Minor discrepancies between a taxpayer’s filed return and third-party data held by the CRA — such as slips from financial institutions, employers, or cryptocurrency platforms — may now be sufficient to trigger automated review and, ultimately, a human-confirmed tax audit. Complete and accurate reporting, reconciled against all available slips, is no longer merely good practice — it is a risk management imperative.
• Cryptocurrency and digital asset holders face a materially elevated tax audit risk. AI-driven analysis has made unreported cryptocurrency transactions significantly easier for the CRA to detect. Canadian taxpayers who have failed to report cryptocurrency income, capital gains, or dispositions face a meaningfully increased risk of CRA identification and tax reassessment.
• Self-employed individuals and small business owners should scrutinize expense claims. AI cross-referencing of reported expenses with industry norms and third-party data means that outlier deductions are more likely to be flagged than in prior years. Taxpayers in high-deduction industries should maintain detailed contemporaneous records.
• The “human oversight” safeguard is real but limited. While the CRA maintains that humans make final decisions, an AI-flagged file is substantially more likely to proceed to tax audit than one that is not. The upstream role of AI in the process matters, regardless of where formal decision-making authority nominally rests.
• There is currently no clear mechanism for taxpayers to challenge algorithmic tax audit selection. Current procedural frameworks do not provide a straightforward pathway for taxpayers to discover or challenge the role AI played in identifying their return for tax audit.

The practical consequence is that a taxpayer flagged in error by an algorithm has no front-end remedy. Defence must be mounted after the audit begins, at the taxpayer’s expense. This is a developing area of Canadian administrative law that experienced Canadian tax lawyers are monitoring actively.

David J Rotfleisch, CPA, JD is the founding tax lawyer of Taxpage.com and Rotfleisch & Samulovitch P.C., a Toronto-based boutique tax law corporate law firm and is a Certified Specialist in Taxation Law who has completed the CICA in-depth tax planning course. He appears regularly in print, radio and TV and blogs extensively.  

With over 30 years of experience as both a lawyer and chartered professional accountant, he has helped start-up businesses, cryptocurrency traders, resident and non-resident business owners and corporations with their tax planning, with will and estate planning, voluntary disclosures and tax dispute resolution including tax audit representation and tax litigation. Visit www.Taxpage.com and email David at david@taxpage.com.

Read the original article in full on Tax Law Canada. Author photo courtesy Rotfleisch & Samulovitch P.C. The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances. Title image: iStock photo ID 1050855182.