On June 24, 2026, the IRS Office of Professional Responsibility (OPR) released its first guidance on how Circular 230 applies to artificial intelligence in tax practice. If you were waiting for a dramatic new rulebook, you did not get one. What OPR delivered instead was something more practical: the standards we already live by do not bend for new technology.
AI does not get its own carve-out. It gets held to the same bar we do.
That may sound like a non-answer, but it is actually the answer. A lot of the anxiety in our profession right now comes from waiting for permission. Some of us want permission to use these tools, and some of us want permission to keep ignoring them. OPR gave us neither. It told us to be professionals.
Let’s be clear about what this guidance does not say, because the fear-based reading is already going around. OPR is not telling practitioners to avoid AI. In fact, the guidance acknowledges what firms already know, these tools are already in our workflows. They are embedded in tax research platforms, document review software, and the automation systems many of us use every day, sometimes without realizing AI is under the hood. Pretending otherwise is not caution. It is denial.
What OPR is saying is that AI must be used with human oversight, professional judgment, due diligence, confidentiality safeguards, and accountability. There is nothing in that list we did not already owe our clients before any of these tools existed. The guidance simply names the obvious point that efficiency gains do not come with a discount on responsibility.
The substance of the guidance maps AI use onto rules we have had for years. Due diligence under Section 10.22 still means verifying facts, calculations, and citations. Now that includes anything an AI system gives you. Competence under Section 10.35 still requires knowledge, skill, thoroughness, and preparation necessary for the matter. But competence now also means understanding the tools you are using well enough to recognize their limits. Written advice under Section 10.37 must rest on verified facts and applicable law, no matter who or what produced the first draft. Firm oversight under Section 10.36 means partners are on the hook for establishing policies that keep the practice compliant. That now includes AI policies, training, vendor review, and documentation.
Here is the part to pay attention to: the guidance is explicitly introductory, and it does not create a safe harbor. There is no "the AI did it" defense.
Courts have already sanctioned lawyers for filing fabricated citations that a chatbot invented and no human bothered to check. OPR is signaling, plainly, that these risks are arriving in the tax space too. When they do, the practitioner will answer for the work product, not the vendor and not the model.
There is a specific billing point firms should not miss, and it has real implications under Section 10.27(a). This rule forbids charging unconscionable fees for any matter before the IRS, and the arrival of AI changes that calculation. If these tools slash the time required for research or drafting, the practice of billing for manual labor that was never performed becomes a significant risk. Claiming hours that were not actually spent or double-dipping on an automated task creates exactly the kind of fact pattern that invites regulatory scrutiny.
The standard here is one we should embrace: efficiency gains should benefit the client, not just pad the firm's margin. Transparency is the best safeguard. Firms should be open about their use of AI-assisted workflows and ensure that the resulting time savings are reflected in the final invoice. When fees track the actual value and effort delivered, they remain defensible. The impact of this guidance likely depends on your current fee structure. Firms that bill by the hour are the most exposed, because they cannot bill for hours a tool saved and the gap between time worked and time billed is exactly what draws questions. Firms on a flat project or engagement rate have more room, but transparency still matters, and passing real efficiency back to clients over time is both the right call and the competitive one. Accrual is priced per return rather than by the client's hours, so the efficiency a firm gains on each return is theirs to manage and to share with clients as they see fit. Accrual's role is to make every return faster and cleaner to process. Turning that into fair, transparent billing is a judgment call the firm makes, supported by a clear record of the work.
If one had to bet on where trouble shows up first, it would point to two places.
The first is confidentiality. It is genuinely tempting to paste a client's numbers into a free, public tool to get a fast answer. Doing so may violate Sections 6713 and 7216, which carry civil and criminal penalties, and it can trigger discipline under Circular 230 on top of that.
The right instinct is to treat every public AI tool as a leak waiting to happen. Before client data goes into any system, someone at the firm needs to know whether it retains prompts, trains on submitted data, allows vendor access, or stores information outside approved environments. If you cannot answer those questions, the client's information does not go in.
The second is the slow erosion of verification. Hallucinations are the obvious failure mode, but the subtler one is drift, which is output that looks polished, sounds authoritative, and is wrong in ways that are easy to miss.
AI is very good at sounding confident. Confidence is not correctness.
The discipline that protects you is boring and non-negotiable. You confirm that cited authorities exist and are current, that quotations reflect their source, that facts match the client's records, that calculations were independently checked, and that conclusions actually fit the client in front of you.
You treat every output as a draft. The moment it becomes a final work product is the moment a human has verified it.
It would be easy to read all of this as a list of ways to get in trouble. We would flip it around.
The firms that build real AI governance now are not just reducing risk. They are building a durable advantage.
That governance does not need to be mysterious. It should answer a few basic questions: which tools are approved, which are prohibited, what client data can and cannot be used, what level of human review is required, how AI-assisted work is documented, how staff are trained, and who makes the call when the answer is uncertain.
If your firm prepares returns, this should also connect to your written information security plan. AI is now part of the data environment. Treating it as separate from security is the wrong move.
Clear guardrails are what let people move fast without fear. A practitioner who knows exactly which tools are sanctioned and how output must be checked will use AI more confidently, not less, than one operating in a fog of vague dread.
Governance is not the brake on adoption. It is what makes responsible adoption possible.
This is exactly the problem we designed Accrual to solve. Instead of sending client information through public AI tools, firms work inside a controlled environment with clear policies around data handling, model training, and auditability. AI can help move the return forward, but it does not turn the work into a black box. Every output is designed to be reviewed by a human before it becomes final, with a clear record of what was generated and what was verified. Accrual manages the underlying technology and security infrastructure, but the firm retains the professional judgment and ultimate accountability Circular 230 has always required.
Strip away the section numbers and the guidance reduces to one idea: AI can support your work, but you remain accountable for it.
It can draft, summarize, organize, and accelerate. It cannot exercise judgment, and it cannot be disciplined by the IRS.
You can.
That is not a limitation to resent. It is a description of what our profession has always been. OPR just confirmed that the arrival of a powerful new tool does not change who signs the return.
This post is part of Accrual's CPA Series, where we explore the technical accounting challenges behind AI-powered tax preparation.