Artificial intelligence (AI) has infiltrated every part of modern life. It’s in our phones, our cars, and even in some of our refrigerators. The tax industry is no different. Over just the past few years, new AI products have been popping up almost daily, promising to make us more efficient and productive. These products range from research tools to review programs to full-on tax return preparation.
But how much can we really rely on these amazing new systems? And how safe is it to dump client data into them?
AI can do some very impressive things, and it’s tempting to believe that it will make our lives much easier; it might. But especially at this stage of development, it’s important to understand the technology’s limitations, as well as the ethical obligations involved in its use.
The Wild, Wild, Westworld
Technology moves faster than the law, and it can take years to get any real guidance on these new, developing systems. There has been a very spirited debate in the tax world about preparer responsibilities around AI. Lacking any real guidance, preparers were left to figure it out for themselves.
OPR Takes a Stand…Sort Of
In late June, 2026, the Office of Professional Responsibility issued OPR Alert 2026-19 addressing some guidelines and the ethical use of AI for tax professionals.
The guidance focuses on some key provisions of Circular 230, and how they relate to tax prep in the age of AI.
Due Diligence
When preparing returns, documents, or other tax-related papers, practitioners must exercise due diligence, per Circular 230 §10.22. Included in this section is the requirement to determine the correctness of oral or written representations made to the Department of the Treasury or the Internal Revenue Service.
This becomes particularly important when we’re dealing with technology that tends to make things up when it doesn’t know the answer. This phenomenon, which has been given the technical term “fabricated outputs” but that we all just call “hallucinating,” can result in the AI models delivering very convincing supports and citations that do not exist.
It should go without saying, but unfortunately doesn’t seem to, that if an AI tool, no matter how sophisticated, delivers a beautifully cited answer to a query, the practitioner should, at the very least, go read those citations. Or, at the VERY least, go find them to confirm that they are real. And then read them.
Hallucinations aren’t the only potential pitfall with AI outputs. In closed systems like Thomson Reuters CoCounsel or Notebook LM, the model is limited to searching references within that system. In open systems like ChatGPT or Claude, the entire internet is available to pull from. This means that if, say, numerous articles had been written and published online about a particular preliminary version of a bill that varied substantially from the bill that was ultimately passed, the AI will grab from all the available info and may deliver results that are no longer accurate.
If you don’t believe us, just take a look at how many of the AI models STILL think that the tips deduction reduces AGI. Some tax-specific AI tools perform better than the general-purpose systems, but even in high-end tax research AI tools, the outputs must still be verified.
Efficiency – But at What Cost?!?
One area of guidance that practitioners may take issue with centers on §10.27(a) – Fees. This section states that a practitioner cannot charge an “unconscionable fee” related to a matter before the IRS.
But what exactly does that mean? What is unconscionable to one client may not be to another. The issue is further complicated by the use of AI. If you’re charging by the hour, how does the time-saving aspect of AI use impact your billing? If a task that used to take 3 hours now takes 10 minutes, do you reduce your fee to reflect that? OPR says yes.
The new guidance suggests that AI-assisted activities should be disclosed to clients and further recommends that “cost reductions” be credited back to the client. This is an obvious concern when billing is done on an hourly basis, but what about practitioners who use value-based (flat fee) pricing? If we bill based on our knowledge and expertise, should we reduce our client’s invoices simply because we can do research a bit faster?
This position also seems to ignore the added cost of using these systems. Most tax-related AI is not free, and if it is, client data should almost certainly not be loaded into it. When it comes to pricing in the age of AI, hourly billing may go the way of AOL installation discs.
Competence
Circular 230 §10.35 requires that a practitioner possess the skill, knowledge, and preparation needed for matters for which they are engaged. We generally think of this in terms of being competent in the area of taxation we are working with. If a preparer is unfamiliar with the rules regarding foreign account reporting, they must either become competent or pass the engagement along to someone who is. In the context of AI, though, the requirement to have knowledge and skill goes beyond understanding tax matters to understanding, at least somewhat, how the systems work.
If a practitioner doesn’t understand that OpenAI models train on the data they receive, they might not worry about uploading sensitive client information. A preparer who is aware of the way these models operate will, hopefully, be more mindful and use an enterprise-level plan that offers higher levels of data protection. Knowing where to find privacy settings and what the limits are is vital to protecting client data.
It is also important that practitioners understand that there are limits to what AI can do and that its outputs may contain bias or inaccuracies. This adds a whole new level to Circular 230 competence.
Systems for the Systems
§10.36 of Circular 230 requires that a practitioner who implements these systems in their practice have policies to ensure compliance for their use. This means that staff must be trained in the proper use of these new tools, and that all tools be vetted.
The Written Word
When it comes to written advice, Circular 230 §10.37 lays out some guidelines that practitioners must follow. The new OPR guidance explains that when using AI to generate written advice, the outputs cannot be relied upon without independent verification. It can be tempting to look at the beautifully formatted documents that AI tools can deliver and pass them along without more than a cursory glance. If we do that, we risk violating Circular 230.
Full Disclosure
Lastly, OPR addresses whether the use of AI tools would require signed disclosures from clients. This topic has been hotly debated among tax professionals and will likely continue to be, since OPR’s stance is, well, lukewarm: “GAI platforms may present risks regarding the unauthorized disclosure of sensitive taxpayer information, especially when data is uploaded to unsecured or public systems.”
It’s important to note that there is an auxiliary services exception to the disclosure requirements under IRC §7216 (the code section governing disclosure and requiring client authorization). That’s why we don’t have to tell clients that we are disclosing their data to Intuit or Drake when we use tax software to file their returns. Whether or not certain AI tools fall under this auxiliary services exception is a grey area.
For instance, using a dedicated AI data input tool when preparing returns seems auxiliary to filing the return (some tax software already offers document-upload data input). Dumping a client return into ChatGPT or Claude likely would not qualify under this exception and would require disclosure (or simply redacting the data).
We believe it is a best practice to get §7216 disclosures from clients when using AI in practice. It is easy enough to provide this at the same time that the engagement letters are provided, and it will save a lot of worry and headache down the road when AI becomes a part of everything we do. We will likely do a deep dive into this topic with samples at our upcoming Tax Update for 2026 Returns seminar.
The End…Or Is It?
This is not likely to be the final word on the ethical use of AI in the tax industry. As technology becomes increasingly prevalent, the rules will undoubtedly shift to reflect the times. Until then, we cannot ignore the implications of Circular 230, IRC §7216, and this new guidance.