Diligence, Privilege, and the Signature at the Bottom

The professional standard that governs this question in United States practice predates the technology and applies to it without amendment.

Circular 230, published as Title 31 Part 10 of the Code of Federal Regulations, governs practice before the Internal Revenue Service by attorneys, certified public accountants, enrolled agents and others. Section 10.22 requires a practitioner to exercise due diligence in preparing, approving and filing returns, documents and other papers relating to IRS matters, and in determining the correctness of oral or written representations made to the Department of the Treasury and to clients.

That provision does two things worth noticing.

It attaches diligence to the practitioner, personally, as a condition of practice. It is not a firm-level standard that a process can discharge.

And it covers representations made to clients, not only to the Service. Advice given to a client that was not diligently determined is within scope even if nothing was ever filed.

Circular 230 also addresses reliance on others. A practitioner may generally rely on the work product of another person if reasonable care was used in engaging, supervising, training and evaluating that person. The framing is instructive: reliance is permitted where the practitioner has taken care about the source and remains responsible for the result.

Whether a software tool is another person for this purpose is a question the text was not written to answer. But the direction it establishes is clear enough for practice. Care in choosing the tool, care in supervising its output, and responsibility that stays with you.

It is worth being precise about who bears what, because practitioners often misplace the risk.

The taxpayer's exposure. Under section 6662 of the Internal Revenue Code, an accuracy-related penalty of twenty percent of the underpayment can apply, including for a substantial understatement of income tax. The substantial-authority analysis discussed in lesson one is a defence to it, as is adequate disclosure combined with a reasonable basis.

The preparer's exposure. Under section 6694(a), a penalty for an understatement due to an unreasonable position, where an undisclosed position lacked substantial authority or a disclosed position lacked reasonable basis. For positions other than tax shelters and reportable transactions, the amount is the greater of one thousand dollars or fifty percent of the income derived by the preparer from the return. Under section 6694(b), where the understatement is due to willful or reckless conduct, the penalty rises to the greater of five thousand dollars or seventy-five percent of the income derived.

And separately, the practitioner's standing to practise, which Circular 230 governs through the Office of Professional Responsibility, with sanctions up to disbarment from practice before the Service.

What none of these do is contemplate a tool as a defendant or a defence. There is no provision under which relying on a system reduces exposure, and no mechanism by which a vendor absorbs a preparer penalty.

So the calculation for a practitioner is one-sided. The tool can save time. It cannot share risk. That asymmetry should shape how much verification feels proportionate, because the answer is more than it feels like.

Confidentiality in tax practice is frequently overestimated, and the misunderstanding becomes dangerous once client information starts moving between systems.

Section 7525 of the Internal Revenue Code, added by the IRS Restructuring and Reform Act of 1998, extends the common law protections of confidentiality that apply to communications between a taxpayer and an attorney to communications between a taxpayer and a federally authorized tax practitioner, which includes certified public accountants and enrolled agents.

The limits are substantial and each one matters.

It applies only to tax advice. Return preparation itself is generally not protected, and a great deal of what a tax practice does is return preparation.

It applies only in noncriminal tax matters before the IRS and noncriminal proceedings in federal court brought by or against the United States. It does not apply in a criminal matter, and it does not apply in litigation with a third party.

And it does not apply to written communications in connection with the promotion of a tax shelter.

So a practitioner's file is far more exposed than an attorney's, and material that was assumed protected frequently is not.

The implication for tooling is direct. Anything typed into a system is potentially discoverable, and a candid working exchange about the weaknesses in a client's position is exactly the material that would be most damaging and least likely to be privileged.

That is an argument for care about what goes into any tool, and about which working material is written down at all.

There is a provision here that surprises practitioners who have not looked at it recently, and it constrains tooling choices directly.

Section 7216 of the Internal Revenue Code makes it a criminal offence for a tax return preparer to knowingly or recklessly disclose or use information furnished in connection with preparing a return, other than as permitted. Section 6713 imposes a related civil penalty. The regulations set out permitted disclosures and a consent regime, with specific requirements for the form and timing of taxpayer consent.

The reason this matters for AI adoption is that disclosure to a third party is disclosure, and a service provider is a third party. The regulations do accommodate disclosures to persons providing auxiliary services in connection with preparation, which is the route by which ordinary outsourced processing operates, and there are particular requirements where processing occurs outside the United States.

This is a place to take advice rather than to reason from first principles, because the provision is criminal and the permitted-disclosure rules are detailed.

The practical posture that follows. Tools that handle return information are a procurement decision with a compliance dimension, not an individual choice. The contract matters, the location of processing matters, and whether consent is required matters.

And the everyday failure is the same one that appears throughout this catalogue. An individual practitioner, under deadline, pasting client information into a consumer tool because it was faster. Nothing visible goes wrong, and a firm-level obligation has been engaged by a personal decision nobody recorded.

Assembling the obligations that do not move, against the tasks that do.

On the left, what a tool can carry. Finding candidate authorities. Extracting data from source documents. Structuring and writing a memo whose analysis you supplied. Producing a client explanation of a position you reached. Drafting routine correspondence.

On the right, what stays with the practitioner, and the reason in each case.

Diligence as to accuracy, because Circular 230 attaches it personally as a condition of practice.

Weighing authority into a confidence level, because that determines disclosure and preparer exposure.

The signature, because section 6694 exposure follows the preparer and no provision lets a tool hold it.

The decision about what client information leaves the firm, because section 7216 makes unauthorised disclosure a criminal matter.

And the judgement about what the client actually needs, which is the part no standard mandates and every client is paying for.

The useful observation about this diagram is that the right-hand column is not a list of things models happen to be bad at. It is a list of things the regulation has assigned to a named human, which is a more durable boundary.

The economic question for tax practices is what happens when research time falls, and the answer depends on which part of the practice you are in.

Compliance work has been priced per return or per engagement for a long time, so a reduction in preparation time flows to margin rather than to a smaller invoice. That is a straightforward gain.

Advisory work billed hourly faces the familiar problem: if a memo that took eight hours now takes three, hourly billing converts an efficiency gain into a revenue reduction. That is not a reason to work slowly. It is a reason the pricing model becomes the constraint rather than the capability.

The adjustments practices actually make. Pricing advisory work by the value of the question rather than the hours, which many firms had already begun for other reasons. Taking on the advisory work that was previously uneconomic, since a question that could not justify eight hours can justify three. And moving up the value chain toward planning, which compresses least.

That middle one is worth sitting with. There is a substantial body of tax questions that clients simply do not ask because the fee for answering properly exceeds the value at stake. A lower cost per question does not just make existing work cheaper. It makes previously unaskable questions askable, which is a market expansion rather than a contraction.

What does not change is the risk profile. Faster research at the same standard of diligence is a gain. Faster research at a lower standard of diligence is the same exposure with a shorter file.

Naming what is not substitutable, which in tax is unusually easy to state because the regulation has drawn much of the line already.

The signature. Someone is exposed under section 6694 and someone can be sanctioned under Circular 230, and that person's willingness to sign is the product. A position nobody will sign is not advice.

The judgement about confidence. Whether a set of authorities amounts to substantial authority is a call, informed by how the Service has behaved, how courts have treated similar arguments, and how aggressive this particular taxpayer wants to be. That last input is not in any corpus.

Knowing what the client did not tell you. Clients describe transactions incompletely, in good faith, omitting the fact that changes the answer. Drawing it out is a conversational skill and it is where advice most often goes wrong.

The view across years and entities. Planning is not a sequence of independent questions. It is a shape over time, and each answer constrains the next.

And representation. If a position is examined, someone appears, argues it, and negotiates. That is a relationship with the Service, conducted by a person with standing to practise.

What is genuinely different about tax, compared with the other professions in this catalogue, is how explicit the boundary is. Elsewhere the argument that judgement and accountability remain human is inferred from how the work behaves. Here it is written into the standards, with a penalty attached and a name on the return.

What a practice can write down, derived entirely from obligations that already existed.

Authority. No citation enters a memo, a letter or a return position unless the person signing has read it at source. Not spot-checked. Every one. Generated output is a lead and is never itself cited.

Tools. Research is done on retrieval-based tools that search an actual corpus and link every proposition to a retrievable source. General chat interfaces may be used for framing and drafting, never for finding authority.

Client data. Return information goes only into tools procured by the firm under contracts that address the section 7216 position, including where processing occurs. Personal accounts are not used for client work. Where consent is required, it is obtained in the prescribed form before the disclosure, not after.

Calculation. Numbers come from the tax software or the spreadsheet. A language model explains a number and never produces one.

Memos. The model drafts from your analysis and your read authorities. Review checks three things: each authority characterised as you understood it, contrary authorities still present, and the conclusion not firmed up.

Signature. The person signing has personally satisfied themselves of the position. Circular 230 section 10.22 attaches diligence to them, and no process discharges it.

Six paragraphs, none of which required a new rule to write. That is the recurring finding across regulated professions: the existing standards were drafted at a level of generality that already covers this.