Artificial intelligence has been heavily marketed to the accounting profession as a panacea for the ongoing talent shortage and the crushing weight of tax season compression. But as firms rush to integrate generative AI into their workflows, the Internal Revenue Service has issued a sobering reminder: you can outsource the labor to a machine, but you cannot outsource the liability.
In a decisive move that sets the regulatory tone for the future of tax preparation, the IRS Office of Professional Responsibility (OPR) recently issued guidance clarifying how existing Treasury Circular 230 rules apply to the use of artificial intelligence. The underlying message is unambiguous: the technology may be novel, but the practitioner’s duties of due diligence, competence, and confidentiality remain entirely unchanged.
Drawing a Line in the Silicon: Circular 230 Meets Generative AI
For months, the accounting profession has operated in a gray area regarding how autonomous or semi-autonomous tools interact with regulatory standards. The OPR's latest guidance eliminates that ambiguity. According to the IRS, practitioners cannot use an AI platform's "black box" nature as an excuse for inaccurate filings, missed deadlines, or compromised client data.
The guidance specifically highlights several core tenets of Circular 230 that are most vulnerable to AI-induced errors:
- Section 10.22 (Diligence as to Accuracy): Practitioners must independently verify the output of AI tools. Accepting an AI-generated tax position or calculation without human review constitutes a failure of due diligence.
- Section 10.35 (Competence): A practitioner must possess the necessary competence to engage in tax practice. If a CPA does not understand how an AI tool arrives at its conclusions, relying on it may be deemed a breach of professional competence.
- Section 10.20 (Information to be Furnished): AI hallucinations—where a model confidently invents facts or case law—pose a severe risk when responding to IRS information requests.
"The rapid adoption of AI in tax practice is a double-edged sword. It offers unprecedented efficiency, but it also scales the potential for systemic errors. The OPR is making it clear that the buck stops with the human whose PTIN is on the return."
To visualize the shift in risk management, consider how traditional Circular 230 duties map onto AI-assisted workflows:
| Circular 230 Duty | Traditional Practice Focus | AI-Assisted Practice Risk |
|---|---|---|
| Due Diligence | Reviewing junior staff workpapers and client-provided documents. | Identifying AI "hallucinations" and verifying algorithmic interpretations of tax code. |
| Confidentiality | Securing physical files and using encrypted client portals. | Preventing client data from being ingested into public Large Language Models (LLMs) for training. |
| Competence | Maintaining CPE and understanding current tax law. | Understanding the limitations, biases, and data cut-offs of the specific AI software being utilized. |
The Murrin Warning: Preparer Liability and the Unlimited Statute of Limitations
The IRS's strict stance on AI oversight arrives at a time when the legal consequences for preparer misconduct are being firmly cemented by the highest courts. Recently, the Supreme Court declined to hear an appeal of the Third Circuit's decision in Murrin, effectively letting stand the IRS's power to assess tax at any time in cases of preparer fraud.
Under Internal Revenue Code Sec. 6501(c)(1), the standard three-year statute of limitations for assessing tax is waived in cases involving a false or fraudulent return with the intent to evade tax. The Murrin decision reinforces that the taxpayer's intent is irrelevant if the preparer committed the fraud. The IRS can pursue the innocent taxpayer indefinitely.
While Murrin dealt with intentional human fraud, its implications for AI-driven practices are profound. If a firm deploys an AI tool that systematically applies an overly aggressive, baseless, or fabricated tax position across hundreds of returns, and the firm fails to exercise due diligence to correct it, could that reckless disregard for the truth be construed as preparer fraud? The legal threshold between gross negligence and fraud is notoriously thin. If an unmonitored AI triggers an unlimited assessment window for a firm's entire client roster, the resulting malpractice litigation would be an extinction-level event for the firm.
Navigating Nuance: Where AI Fails and Human Expertise Wins
The necessity of human oversight is most apparent when dealing with evolving, highly technical areas of the tax code where historical training data is insufficient. AI models thrive on established, static patterns. They struggle with transitional rules and sudden legislative shifts.
A prime example is the recent release of transitional guidance on Qualified Opportunity Zones (QOZs) by the IRS. Following recent legislative adjustments, accounting professionals are now tasked with navigating complex transitional rules regarding capital gains deferrals, investment timelines, and compliance reporting for QOZ funds.
An AI model trained on 2024 or 2025 data might confidently apply outdated QOZ regulations, leading to massive compliance failures for high-net-worth clients. The OPR's Circular 230 guidance dictates that practitioners cannot blame the software if a client loses their QOZ tax benefits due to an algorithmic oversight. CPAs must actively bridge the gap between an AI's historical knowledge base and the reality of real-time IRS transitional guidance.
The Legislative Horizon: AI as a Tax Target
Beyond using AI to prepare taxes, the accounting profession must also prepare to advise clients on how AI itself will be taxed. The rapid capitalization of AI companies has drawn the attention of lawmakers looking to capture revenue from the tech boom.
Notably, Senator Bernie Sanders recently proposed a radical remittance tax on "systemically important AI activity." Under this proposal, targeted AI companies would be required to pay a tax amounting to 50% of their outstanding equity interests. While this proposal faces significant legislative hurdles, it signals a broader macroeconomic trend: lawmakers are viewing AI not just as a tool, but as a distinct asset class and revenue source that requires novel taxation mechanisms.
For tax advisory practices, this represents a dual mandate. Firms must rigorously govern their internal use of AI to comply with Circular 230, while simultaneously developing the specialized expertise required to advise tech clients through an increasingly hostile and complex legislative environment.
Actionable Steps for Firm Leaders
To navigate this intersection of regulatory scrutiny and technological advancement, firm leadership should immediately implement the following protocols:
- Establish an AI Acceptable Use Policy (AUP): Draft clear, written guidelines detailing which AI tools are approved for firm use, what client data can be inputted, and the mandatory review processes required before AI output is utilized in client deliverables.
- Mandate "Human-in-the-Loop" Workpapers: Require staff to document their independent verification of AI-generated research or calculations. Workpapers should clearly delineate what was produced by AI and how the human preparer validated it against primary source documents.
- Audit Vendor Agreements: Review the terms of service for all third-party tax software integrating AI. Ensure they comply with Circular 230 confidentiality requirements and do not use client financial data to train public models.
Conclusion: The Premium on Professional Skepticism
The IRS Office of Professional Responsibility has drawn its line in the sand. As AI continues its march toward commoditizing basic compliance work, the true value of the CPA will no longer be in data entry or rote calculation. The premium will be placed squarely on professional skepticism, ethical judgment, and the assumption of liability.
The Supreme Court’s refusal to limit the IRS’s reach in preparer fraud cases serves as a stark reminder of the stakes involved. In the AI era, a firm's most valuable asset is not the sophistication of its technology stack, but the rigor of its quality control. Those who treat AI as an infallible employee do so at their own peril; those who treat it as a powerful, yet inherently flawed tool requiring rigorous human oversight will define the next era of the accounting profession.