Part of our AI Regulation and Compliance News series
Circular 230 and AI: The Tax Practitioner Checklist for IRS OPR Alert 2026-19
On June 24, 2026 the IRS Office of Professional Responsibility issued Alert 2026-19, titled "Introductory Guidelines for Responsible AI Use in Federal Tax Practice." It is the first time the office that enforces Circular 230 has spelled out how those rules apply when a tax professional puts a return, a memo, or a client letter through a generative AI tool. The short version: nothing in Circular 230 changed, and that is exactly the point. The duties you already owe still run, and now they run over your AI output too.
If you are a CPA, an enrolled agent, or a tax attorney, this briefing turns the four duties OPR named into a desk-ready checklist you can stand up before the next filing season. The goal is simple. When you sign a return, the AI work product is still your signature.
Key takeaways
- The output is still yours. Alert 2026-19 confirms that due diligence under Circular 230 section 10.22 cannot be handed to an algorithm. If AI drafted it, you verify every figure, citation, and calculation before it carries your name.
- Competence now includes the tool. Section 10.35 means you have to understand how your AI system works, where it breaks, and what it should never touch. Using a tool you cannot explain is its own exposure.
- Firm procedures are expected, not optional. Section 10.36 points to staff training, an approved-tools list, secure data handling, accuracy checks, and written documentation. A solo or small practice can satisfy this on one page.
- Client data has a hard line. Confidentiality duties bar feeding taxpayer information into unsecured or public AI tools. The safest rule is to never paste client identifiers into a tool that is not covered by a data agreement.
What OPR Alert 2026-19 actually means
OPR is the IRS office that polices practitioner conduct under Treasury Department Circular No. 230. It does not write tax law. It governs how people who practice before the IRS are allowed to behave. So when OPR speaks on AI, it is not telling you which model to buy. It is telling you that the existing standard of care now has a new fact pattern, and that fact pattern is software writing first drafts of your work.
The alert is labeled introductory and is not a new binding rule. That has led some practitioners to wave it off. That read is a mistake. The guidance does not need to create a new rule, because it re-anchors duties that are already enforceable. A practitioner who relies on an AI hallucination, signs a return built on a fabricated citation, or uploads a client Social Security number to a public chatbot is not protected by saying the guidance was only introductory. The underlying Circular 230 sections were always in force. Alert 2026-19 simply removed the excuse of not knowing how they apply.
Reporting from Accounting Today and CPA Practice Advisor on June 24, 2026 frames it the same way: the IRS is signaling expectations early, before AI use becomes universal in tax practice, so that firms build good habits now instead of cleaning up after a problem return.
What the duty actually requires
OPR pointed at four areas. Here is what each one asks of a working practitioner, in plain terms.
Due diligence (section 10.22)
This is the core of it. You must exercise due diligence in preparing and filing, and in any oral or written representation to the IRS or a client. AI does not lower that bar. If a tool drafts a position, you confirm the authority is real and on point. If it runs a calculation, you check the math against the source numbers. A confident, well-formatted answer is not the same as a correct one, and the duty falls on you to tell the difference.
Competence (section 10.35)
You cannot competently use a tool you do not understand. That means knowing, at a working level, what your AI system is good at, where it tends to fail, what data it retains, and which tasks it should never be trusted with unsupervised. Competence also means staying current as the tools change, which they do often.
Firm procedures (section 10.36)
A practitioner with oversight responsibility must take reasonable steps to ensure the firm has adequate procedures. For AI, that reads as staff training, a vetted list of approved tools, secure handling of taxpayer data, a way to monitor accuracy, and documentation that the procedures exist and are followed. A one-person shop is not exempt. You are both the firm and the staff, so the procedure is just written down once and followed.
Confidentiality
Tax practitioners hold some of the most sensitive data a person owns. Pasting that into a public AI tool can expose it and can breach your confidentiality obligations. The guardrail is to treat any tool without a signed data protection agreement as off limits for identifiable client information.
A practical what-to-do-now workflow
You do not need an enterprise compliance department to meet this. You need three short documents and one habit. Here is the order to build them before season.
1. Write a one-page AI acceptable-use and verification rule. State which tools your practice is allowed to use, what client data may and may not go into them, and the rule that every AI-produced figure, citation, and calculation gets human-verified and logged before it reaches a return or a client. This single page covers your section 10.22 and section 10.36 expectations at once.
2. Build an approved-tools register. List each AI tool you use, who reviewed it, whether it has a data protection agreement, and what kinds of work it is cleared for. When you adopt a new tool, it goes on the register before it touches client work, not after.
3. Keep a verification log. For AI-assisted work, note what the tool produced and what a person checked. This does not have to be heavy. A short line per engagement showing a human reviewed the output is the documentation that turns "we use AI carefully" into something you can show.
4. Make verification a step, not a vibe. The habit that matters most is reading every cited authority and re-running every number the tool gives you. Treat AI output as a first draft from a fast junior preparer who is sometimes confidently wrong, and review it exactly that way.
| Safe to delegate to AI (then review) | Stays on the practitioner's desk |
|---|---|
| First-draft client letters and engagement recaps | The final signature and the position taken on the return |
| First-pass memos you will rewrite and verify | Confirming every authority is real and on point |
| Reconciliations and summaries you check against source | Judgment calls on aggressive or uncertain positions |
| Drafting routine information requests | Anything touching identifiable client data in an unvetted tool |
The verification and competence guardrails
Two failures will get a practitioner in trouble faster than any other. The first is the fabricated authority. Generative tools can invent a case, a code section, or a revenue ruling that reads perfectly and does not exist. The fix is non-negotiable: open and read every source the tool cites before you rely on it. The second is the silent calculation error. A tool can carry a wrong figure through a clean-looking schedule. The fix is to re-run the numbers from the source data yourself.
Competence is the quieter guardrail. Before you trust a tool with a category of work, you should be able to say in a sentence how it handles that work and where it tends to fail. If you cannot, you are not yet competent to use it unsupervised on that task, and Circular 230 expects you to know the difference. That same competence discipline is what state boards are now writing into CPA ethics CPE, which we cover in a companion briefing below.
What AI does not replace
The alert is a useful reminder of where the machine stops. AI does not hold the license. It does not sign the return. It does not owe a duty to the IRS or to your client. It cannot exercise judgment about a gray-area position, and it cannot be sanctioned, because it is not a person practicing before the IRS. You are. Every duty Circular 230 names lands on the human in the chair. Used well, AI gives you back hours on drafting and reconciliation so you can spend more of your attention on the judgment that only you can supply. Used carelessly, it just produces wrong work faster, with your name on it.
Build the AI skills the duty now assumes
Circular 230 competence is no longer abstract. You are expected to understand the tools you use. The Leveraged CPA and Finance program teaches accountants and tax pros how to use Claude for drafting, reconciliations, and first-pass memos while keeping verification and client data exactly where the rules require.
Want a lower-cost first step? Join The Leverage Club for $49 for the working sessions, templates, and a room of professionals doing this carefully. Not sure which path fits your practice? Take the two-minute course finder quiz.
- How the IRS Uses AI to Select Returns for Audit. The other side of the desk, where the government points AI at your clients' returns.
- AI and the New CPA Ethics CPE From State Boards. How state-level ethics requirements are catching up to the same competence question.
Frequently asked questions
Does IRS OPR Alert 2026-19 ban tax practitioners from using AI?
No. The June 24, 2026 alert does not prohibit AI. It confirms that existing Circular 230 duties apply when you use it. You may use AI for drafting and analysis as long as you verify the output, understand the tool, keep firm procedures, and protect client confidentiality.
Is the AI work product still my responsibility if a tool produced it?
Yes. Under Circular 230 section 10.22, due diligence cannot be delegated to software. Every figure, citation, and calculation an AI tool produces must be human-verified before it reaches a return or a client. When you sign, the output is yours.
Can I put client tax data into a public AI chatbot?
You should not. Confidentiality duties bar feeding taxpayer information into unsecured or public AI tools. Use only tools covered by a data protection agreement, and keep identifiable client data out of any tool you have not vetted for security.
Sources: IRS Office of Professional Responsibility, Alert 2026-19, "Introductory Guidelines for Responsible AI Use in Federal Tax Practice," June 24, 2026. Accounting Today, "IRS rolls out introductory AI guidelines for tax practitioners," June 24, 2026 (https://www.accountingtoday.com/news/irs-rolls-out-introductory-ai-guidelines-for-tax-practitioners). CPA Practice Advisor, "IRS Sets Introductory Guidelines on AI Use for Tax Professionals," June 24, 2026 (https://www.cpapracticeadvisor.com/2026/06/24/irs-sets-introductory-guidelines-on-ai-use-for-tax-professionals/185619/).