MYCPE ONE

On June 24, 2026, the IRS Office of Professional Responsibility (OPR) issued its first formal guidance on artificial intelligence — OPR Alert 2026-19, "Introductory Guidelines for Responsible AI Use in Federal Tax Practice."

If you skim it, it sounds almost anticlimactic: no new law, no new form, no new penalty regime. But that's exactly what makes it important. The OPR isn't creating a new rulebook for AI — it's confirming that the old rulebook, Circular 230, already covers it. Which means every CPA, EA, and tax attorney using AI today is already on the hook for how it's used, whether they've thought about it or not.

Here's the plain-English version of what changed, why it matters, and what to actually do about it.

The One-Line Summary

If you sign it, you own it — even if AI wrote most of it.

AI can draft, research, and summarize. It cannot take responsibility. That responsibility sits with the practitioner, full stop.

A Quick Refresher: What Is Circular 230, Anyway?

For anyone who hasn't touched it since exam prep — Circular 230 (31 C.F.R. Part 10) is the set of conduct rules that governs everyone who practices before the IRS: CPAs, enrolled agents, and attorneys. It's been the backbone of practitioner discipline for decades, covering things like:

  • How carefully you have to check your own work (due diligence)
  • Whether you're actually qualified to handle a matter (competence)
  • What you can charge for (fees)
  • How you protect client information (confidentiality)

The OPR is the body that enforces it, and it can suspend or bar practitioners who violate it. Nothing about that structure has changed. What OPR did with Alert 2026-19 is point at six existing sections of Circular 230 and say, plainly: these apply to your AI use too.

Why This Alert, Why Now: A Cautionary Tale Worth Knowing

OPR didn't issue this in a vacuum. It's responding to a pattern that's already played out in the legal and consulting world — most notably a widely reported 2025 incident involving a major consulting firm's Australian office. That firm delivered a lengthy government report that turned out to contain fabricated quotes attributed to a real judge, references to sources that didn't exist, and citations credited to the wrong authors. 

The culprit: generative AI content that nobody had fully verified before it went out the door. The firm ended up refunding part of its fee.

Swap "government report" for "tax memo" or "IRS response letter," and you can see exactly why OPR is drawing this line now, before a similar story involves a tax practitioner's signature instead of a consulting firm's letterhead.

The Six Circular 230 Duties AI Doesn't Let You Off The Hook For

OPR mapped AI risk onto six areas practitioners already have obligations under. Here's each one, translated into plain terms, with a realistic scenario attached.

The Six Circular 230 Duties AI Doesn't Let You Off The Hook For

1. Due Diligence: § 10.22

Plain English: You have to actually check the facts, citations, and math AI gives you — every time — before it reaches a client or the IRS.

Example: An associate asks an AI tool to draft a client memo on a deduction issue. The AI cites a revenue ruling that sounds right and reads convincingly — except it was superseded two years ago. Nobody catches it before the memo goes out. If the IRS later challenges the position, "the AI said so" is not a defense. The practitioner who signed off owns the error.

The takeaway: Treat every AI-generated citation, number, and fact the way you'd treat a first-year associate's first draft — a starting point, not a finished product.

2. Competence and Technological Literacy — § 10.35

Plain English: Competence now includes understanding the tool you're using, not just the tax law.

You don't need to know how a large language model works under the hood, but you do need to understand — at a working level — how it generates answers, where it tends to go wrong (older data, confident-sounding wrong answers, missing context about your specific client), and when its output simply isn't reliable enough to use.

The takeaway: "I didn't realize the AI could just make things up" is no longer a viable excuse under Circular 230. Not understanding your tools is now itself a competence problem.

3. Firm Supervision: § 10.36

Plain English: This one isn't on individual staff — it's on firm leadership.

If a partner or manager lets staff use consumer AI tools without training, policy, or oversight, and something goes wrong, that's not just an individual failure — it's a firm-level compliance gap. OPR expects firms to:

  • Train staff on both the capabilities and the risks of AI tools
  • Document protocols for how client data is handled and how AI accuracy gets checked
  • Vet any third-party or outsourced AI tool before it touches client work

The takeaway: "We never got around to writing a policy" will not be a comfortable sentence to say in front of an examiner.

4. Written Advice: § 10.37

Plain English: Advice AI helps draft still has to rest on facts and assumptions you've actually verified.

If an AI tool's reasoning is essentially a black box — you can't see how it got to its answer — relying on that output without independent verification may itself count as unreasonable reliance under this section. That's a meaningful shift in framing: opacity isn't neutral, it's a risk factor.

The takeaway: If you can't explain how a conclusion was reached, you're not ready to put your name on it.

5. Fees: § 10.27(a)

Plain English: If AI saves you time, that savings needs to show up for the client — not just your margin.

Circular 230 has long banned "unconscionable fees." OPR's new angle: billing a client for three hours of manual research when AI actually did it in twenty minutes could cross that line — especially if it becomes a pattern. Double-billing (charging the AI-assisted rate, then billing again as though the work were done manually) is called out explicitly as a red flag.

Example: A firm uses AI to cut return-prep research time by 70% but keeps billing at the old hourly estimate without disclosure. One instance might be an oversight. A consistent pattern across engagements is the kind of thing OPR says could constitute a violation.

The takeaway: This isn't just a compliance risk — it's a client-trust opportunity. Firms that are transparent about AI-driven efficiency and adjust billing accordingly build credibility instead of exposure.

6. Confidentiality: IRC §§ 6713 and 7216(a)

Plain English: Client tax data can only go into secure, enterprise-approved AI tools — never public or consumer-grade platforms.

Pasting a client's income details, SSN, or return data into a general-purpose public chatbot to "just draft something quickly" isn't a shortcut — it's a potential unauthorized disclosure with civil and criminal exposure under §§ 6713 and 7216(a). Where §7216-protected information is involved, specific signed client consent is typically required before that data touches any AI tool at all.

The takeaway: The tool matters as much as the task. "Enterprise-approved" isn't a nice-to-have; it's the line between compliant and exposed.

Before Your Next AI-Assisted Filing: A Five-Question Self-Audit

OPR's guidance boils down to five questions worth asking on every engagement where AI played a role:

  • Have I independently verified every fact, citation, and calculation AI produced?
  • Do I actually understand how this AI tool generates its answers and where it tends to fail?
  • Is client data only touching secure, enterprise-approved systems?
  • Does my firm have documented AI training and data-handling protocols in place?
  • Does my billing honestly reflect the efficiency AI created?

If you can answer "yes" to all five with a straight face, you're in solid shape. If any of them make you pause, that pause is the work you need to do next.

What Firm Leaders Should Actually Do This Quarter

This guidance rewards firms that act early and quietly, not the ones scrambling after an examiner asks a hard question. A few concrete steps:

  • Write down your review process. Who checks AI-drafted citations? Who verifies AI-generated math? Put it on paper, even if it's short.
  • Audit recent billing. Pull a handful of engagements where AI materially cut research or drafting time. Did the fee reflect that? If not, decide whether prospective adjustment or client disclosure is warranted.
  • Vet your tools. Know which AI platforms your team is actually using — including the ones nobody officially approved. Consumer tools used informally are a firm-level liability, not a personal one.
  • Train, then document the training. A policy nobody's seen isn't a policy OPR will credit you for having.

The Bottom Line

OPR Alert 2026-19 doesn't ban AI, slow it down, or create a new licensing hurdle. It does something arguably more consequential: it removes the ambiguity. The duties of due diligence, competence, supervision, reasonable reliance, fair billing, and confidentiality were never suspended just because a machine did part of the work and now that's in writing, directly from the body that disciplines practitioners.

The firms that treat this as a five-minute policy update will be the ones with clean files when the question eventually comes up. The ones that treat it as an afterthought will be explaining, after the fact, decisions that were never written down in the first place.

AI is a tool. A very good one. But under Circular 230, it has always been and remains — a tool that works for a professional, not as one.

This article summarizes and simplifies OPR Alert 2026-19 (June 24, 2026) and related Circular 230 provisions for general informational purposes. It does not constitute legal or professional advice — practitioners should review the full text of the alert and consult their own compliance counsel as needed.
Amrit Singh

Amrit Singh

Amrit Singh is a business leader with 10+ years of experience in continuing education. Helping accounting, tax, and finance professionals stay compliant with ease, he began his journey as a consultant. Learning across industries before stepping into a leadership role, he is shaped by both successes and failures. Amrit is passionate about problem-solving, building products, exploring technology, and mentoring future leaders. He is dedicated to transform continuing education, making it simpler, smarter, and more meaningful. Through his blogs and talks, he shares insights on accounting careers, CPA compliance, and the future of continuing education.

Must Read Blogs

How AI Is Changing CPE Requirements for Accounting Professionals
  • 306 Reads

How AI Is Changing CPE Requirements for Accounting Professionals

AI CPE Courses vs Traditional Accounting CPE: Which Delivers More Value?
  • 226 Reads

AI CPE Courses vs Traditional Accounting CPE: Which Delivers More Value?

Best AI Certifications for CPAs (2026)
  • 359 Reads

Best AI Certifications for CPAs (2026)

View All