AI in finance and accounting
Section 10.22 requires a practitioner to exercise due diligence in preparing and filing returns and other papers. Paragraph (b) supplies a presumption of diligence for reliance on "the work product of another person" — and a model is not a person. Circular 230 presumes nothing for reliance on machine output.
§ 1When it applies
26 September 2007
Paragraph (a), the diligence duty itself: applicable to conduct on or after this date.
12 June 2014
Paragraph (b), the reliance presumption, applicable from the publication of the current revision of Circular 230 (Rev. 6-2014). One section, two application dates — and it is the later paragraph that a practitioner relying on a tool would want to invoke.
§ 2In practice
The section has two paragraphs and they do different work. Paragraph (a) imposes the duty: due diligence in preparing or assisting in the preparation of, approving and filing returns, documents, affidavits and other papers relating to Internal Revenue Service matters, and in determining the correctness of oral or written representations made to the Treasury Department and to clients. Paragraph (b) then softens it in one defined situation. "Except as modified by §§10.34 and 10.37, a practitioner will be presumed to have exercised due diligence for purposes of this section if the practitioner relies on the work product of another person and the practitioner used reasonable care in engaging, supervising, training, and evaluating the person, taking proper account of the nature of the relationship between the practitioner and the person."
Every operative term in that presumption is addressed to a human being. Engaging, supervising, training and evaluating are things done to people, and "the nature of the relationship between the practitioner and the person" is a professional relationship, not a licence agreement. Read against a language model the words either have no referent or acquire a technical sense the drafters plainly did not intend — a practitioner does not train the model, and evaluating it is not the same act as evaluating a preparer. The presumption is therefore unavailable, and paragraph (a) sits on the practitioner undiluted.
The consequence is a split that most practices have not noticed. A practitioner who passes a draft return to a junior and reviews it has a presumption available, provided the engagement, supervision and evaluation of that junior were carried out with reasonable care. A practitioner who passes the same work to a tool has no presumption at all and must show the diligence directly, on the return in front of them. That is a heavier record, not a lighter one, and it runs in the opposite direction to the reason the tool was bought. The ethics code reaches a comparable conclusion by a different route, requiring a judgement about what verification a given use of technology needs rather than allowing a standing assumption: see using the output of technology.
Two boundaries keep the page honest. Paragraph (b) opens "Except as modified by §§10.34 and 10.37", so written advice is governed separately and the presumption cannot be borrowed across into it. And Circular 230 binds a practitioner in practice before the Service; a controller who never files with it is outside this section entirely, whatever their tooling. Where the same person does both — a finance role in the morning and a return in the afternoon — the diligence standard changes with the work, which is a distinction worth writing into a firm's own procedures rather than leaving to recollection. Competence for the tool itself is a separate provision again: see competence to use an AI tool.
§ 3Who it binds
United States only, and only for practice before the Internal Revenue Service. Circular 230 binds a practitioner — an attorney, certified public accountant, enrolled agent or other person authorised to practise before the Service — in preparing, approving and filing returns, documents, affidavits and other papers, and in the correctness of representations made to the Treasury Department and to clients. A finance function that files nothing with the Service is outside it.
§ 4What discharges it
The artefacts an assessor asks to see, and what makes each one sufficient rather than merely present.
01
It records what was checked rather than what was assumed. The absence of a presumption means the file has to carry the work, and a note saying the output "was reviewed" carries none of it.
02
For the items that were traced, the document and the person; for those that were not, the reason. This is the artefact that distinguishes verification from a plausibility read.
03
Extraction, computation, drafting or research each carry a different exposure, and a practice that cannot say which of them happened cannot scope its own diligence.
04
Supplier accuracy figures are drawn from someone else's inputs. An internal sample, however small, is the difference between a claim relied on and a claim adopted.
05
Re-performance on a stated sample, signed and dated, is the strongest single item in the file. A sign-off that names no sample records an opinion about the work rather than an examination of it.
§ 5Worked example
A tax practice uses a commercial assistant to extract figures from client-supplied documents and populate a return schedule. The engagement partner reviews the completed return in the usual way: a reasonableness read, a comparison against last year, and a check of the totals. The practice has a written policy requiring the assistant to be used only on documents the client has confirmed complete, and it retains the assistant supplier's accuracy claims on file.
Has the partner exercised the due diligence §10.22(a) requires?
Possibly, but nothing in the file shows it. The review described is a review of the return, not of the extraction, and the two are different acts: comparing this year with last will catch a figure that moved and will not catch a figure that was misread consistently in both. The supplier's accuracy claims are marketing material until the practice has tested them on its own document population, and the usage policy governs when the tool may be pointed at something rather than what happens to what it produces. What paragraph (a) wants is evidence that the correctness of the representations on the return was determined — for the AI-extracted figures, that means naming which items were traced back to the source document and why the rest were not. If the partner is treating the tool as a supervised junior and relying on the paragraph (b) presumption to close that gap, the presumption is not there to be relied on.
§ 6What a weak answer looks like
The practice explains that it supervises the software the way it supervises a first-year associate, and points to a review policy that makes no distinction between the two. The comparison is the failure. Supervision is one of four things paragraph (b) asks about, and the presumption it unlocks is available only for the work product of a person. A firm resting on it has adopted a defence that does not exist, and will discover so at the point it most needs the paragraph — after an error has already reached a filed return.
§ 7Exposure
Censure, suspension or disbarment from practice before the Internal Revenue Service, and monetary penalties
Circular 230 is issued under 31 U.S.C. 330, which authorises the Secretary of the Treasury to suspend or disbar a representative who is incompetent or disreputable or who violates the regulations, and to impose a monetary penalty. The consequence is therefore practice sanction rather than a fine on an employer, and it reaches only a person who practises before the Service. Proceedings are brought by the Office of Professional Responsibility.
§ 8Elsewhere
Where another instrument addresses the same obligation. These are correspondences, not comparisons — the Council does not rank one framework against another.
Agentic AI guidance
The agentic material sets out what a user remains answerable for when a tool acts on their behalf. Section 10.22 answers the same question inside a regime that extends its presumption of diligence only to reliance on a person.
A correspondence indicates that two instruments address the same underlying obligation. It is not a mapping endorsed by either body, not a statement that one satisfies the other, and not a judgement about which is more demanding.
§ 9Where this is assessed
Examined in one credential, in the domains named on each card.
Verifying AI output before reliance · 20% of the paper
For finance and accounting professionals who use AI in analysis, close and reporting work — and verify the output before the numbers are relied on.
§ 10provenance
This page sets out what the instrument requires and what discharges it. The official text is the authority — these go straight to it.
§ 11Also read
Certification
What this page sets out is examined, on work the candidate has not seen, in the credential beside it. Every paper is machine marked against its key and every result is reviewed and released by a named person, and every credential is on the public register.
AIAC-FNAF
Certified AI in Finance & Accounting, Fundamentals2 hours · remote, proctored · valid 2 years
Ten questions on using AI at work. About two minutes, free. You finish as one of six types, with a card to share.
SHARE BY 31 DECEMBER, US$199 FEE WAIVED
A free quiz from the AI Assurance Council. Not a certification.