Skip to content
UNTIL 31 DECWhich kind of AI user are you?
Take the two-minute check
AIAC AI ASSURANCE COUNCIL™

AI in finance and accounting

31 CFR §10.37 — Requirements for written advice

Section 10.37 sets what a practitioner must do before giving written advice on a federal tax matter: base it on reasonable assumptions, consider the relevant facts and circumstances, make reasonable efforts to ascertain them, and place no unreasonable reliance on representations. A fabricated authority in a model-drafted memo fails the first and the last together.

§ 1In practice

The four requirements a fluent draft does not meet

Four of the six requirements in §10.37(a)(2) bear directly on a memorandum a model drafted. The practitioner must "Base the written advice on reasonable factual and legal assumptions (including assumptions as to future events)"; "Reasonably consider all relevant facts and circumstances that the practitioner knows or reasonably should know"; "Use reasonable efforts to identify and ascertain the facts relevant to written advice on each Federal tax matter"; and "Not rely upon representations, statements, findings, or agreements (including projections, financial forecasts, or appraisals) of the taxpayer or any other person if reliance on them would be unreasonable". Each is a requirement about the work behind the words, and a fluent draft satisfies none of them by being fluent.

Paragraph (a)(3) defines when reliance becomes unreasonable: where the practitioner knows or reasonably should know that one or more representations, or one or more assumptions on which any representation is based, are "incorrect, incomplete, or inconsistent". That is the paragraph a fabricated citation runs into. Advice resting on an authority that does not say what the memorandum says it says is advice resting on an assumption that is incorrect, and the practitioner is the person who reasonably should have known, because verifying an authority is within their competence and takes minutes. The same problem is why the diligence presumption in §10.22(b) offers nothing here either — see the reliance safe harbour does not cover a model.

Paragraph (b) closes the circle. It permits a practitioner to rely on "the advice of another person" only where the advice was reasonable and the reliance in good faith, and it lists the conditions under which that reliance is not reasonable. A model is not another person, so the paragraph offers no shelter, and the practitioner is answerable for the reasoning as though they had written it. Note also which of the four requirements the parenthesis in the fourth reaches: "projections, financial forecasts, or appraisals". A model-generated forecast pasted into a memorandum is exactly the kind of representation the requirement contemplates, and the discipline for producing one is set out at preparing information with AI assistance.

§ 2What discharges it

What the advice file must contain

The artefacts an assessor asks to see, and what makes each one sufficient rather than merely present.

  1. 01

    An authority verification log for every citation in the advice

    Each authority checked against the source, by whom and when — including the ones that proved sound. A log listing only corrections cannot show that the unlisted citations were looked at.

  2. 02

    The assumption reasonableness note under §10.37(a)(2)(i)

    Assumptions stated separately from conclusions, with the basis for each, including assumptions about future events. Assumptions buried in narrative cannot be tested by a reader or by the practitioner.

  3. 03

    The fact-ascertainment record under §10.37(a)(2)(iii)

    What was asked of the client, what was received and what remained unknown when the advice went out. Reasonable efforts is a standard about enquiry, and enquiry leaves traces.

  4. 04

    A drafting record identifying which passages the model produced

    Not for disclosure to the client, but so the practice can see where its own reasoning stopped and generated text began. Practices that keep no such record cannot review their own exposure.

§ 3Who it binds

Which advice, and whose

Tax practitioner

United States only. The section applies to written advice on a federal tax matter given by a practitioner in practice before the Internal Revenue Service. It attaches to the advice rather than to the engagement, so a matter opened before the section applied is inside it for anything written now, and it reaches any written communication meeting the description regardless of its length, format or whether a fee was charged for it.

§ 4What a weak answer looks like

Checking that a citation exists

"We check the cites." Pressed, the practice means that a reviewer confirmed each cited authority exists. That catches an invented case and misses the commoner failure, which is a real authority cited for a proposition it does not support — the form of error a language model produces most fluently, because the citation is drawn from a genuine neighbourhood of the argument. Checking existence is a search; checking support is reading. The requirement is about the assumptions the advice rests on, and only the second of those acts touches them.

§ 5When it applies

Attached to the advice, not the engagement

  1. 12 June 2014

    Applicable to written advice rendered after this date, the publication of the current revision of Circular 230 (Rev. 6-2014). Because the section attaches to the advice, the relevant date is when the memorandum was written, not when the client was taken on.

§ 6Exposure

Practice sanction

Censure, suspension or disbarment from practice before the Internal Revenue Service, and monetary penalties

Sanction under 31 U.S.C. 330(b), applied through Circular 230 and brought by the Office of Professional Responsibility. The consequence is to the practitioner's ability to practise before the Service rather than a fine on the client or the firm's revenue, and it reaches nobody who does not practise before the Service.

§ 7Elsewhere

Disclosure, treated elsewhere

Where another instrument addresses the same obligation. These are correspondences, not comparisons — the Council does not rank one framework against another.

  • EU AI Act

    Article 50 addresses when the artificial origin of content must be disclosed to the person receiving it. This section says nothing about disclosure and attaches its requirements to the reasoning behind the advice, whatever produced the words.

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.

§ 8Where this is assessed

Where this is assessed

Examined in one credential, in the domains named on each card.

AIAC-FNAFPractice

Certified AI in Finance & Accounting, Fundamentals

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.

Financial servicesEnrolment open

§ 9provenance

The provision itself

This page sets out what the instrument requires and what discharges it. The official text is the authority — these go straight to it.

§ 10Also read

Also read

Certification

Examined on this standard of evidence

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, Fundamentals

2 hours · remote, proctored · valid 2 years

Enrol