← Back to the Library
Library · Guide

Lesson 8: Circular 230

← Study Guide home · Lesson 8 of 12


Treasury's rules for practice before the IRS


Learning Objectives

After completing this lesson, you will be able to:


Who Circular 230 Applies To

Treasury Department Circular No. 230 (31 C.F.R. Part 10) governs practice before the IRS. It applies to:

If you have a PTIN and are licensed or enrolled to practice before the IRS, Circular 230 applies to you.

§10.22: Due Diligence

Full text (key provisions):

"(a) In general. A practitioner must exercise due diligence, (1) In preparing or assisting in the preparation of, approving, and filing tax returns, documents, affidavits, and other papers relating to Internal Revenue Service matters; (2) In determining the correctness of oral or written representations made by the practitioner to the Department of the Treasury; and (3) In determining the correctness of oral or written representations made by the practitioner to clients with reference to any matter administered by the Internal Revenue Service.

(b) Reliance on others. 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..."

Application to AI:

The IRS's Office of Professional Responsibility has specifically addressed AI and due diligence. Key points from the OPR's 2024 National Tax Forum presentation:

The "reasonable care" standard for AI: Under §10.22(b), reliance on another person's work product (which by analogy includes AI output) is presumed due diligent if the practitioner:

  1. Used reasonable care in engaging the tool (selected an appropriate tool for the task)
  2. Used reasonable care in supervising the tool's use (reviewed outputs)
  3. Used reasonable care in training relevant personnel (staff using the AI tool understand its limitations)
  4. Used reasonable care in evaluating the output (did not simply accept AI results without review)

COMPLIANCE NOTE: The §10.22(b) "reliance on others" framework is the closest Circular 230 comes to an AI-specific due diligence standard. Courts and the OPR apply it by analogy. A practitioner who can demonstrate they: (1) selected a tool appropriate for the task, (2) trained staff on its limitations, (3) reviewed AI outputs against primary sources, and (4) documented the review has a strong §10.22 compliance position.

§10.35: Competence

Current text:

"A practitioner must possess the necessary competence to engage in practice before the Internal Revenue Service. Competent practice requires the appropriate level of knowledge, skill, thoroughness, and preparation necessary for the matter for which the practitioner is engaged. A practitioner may become competent for the matter for which the practitioner has been engaged through various methods, such as consulting with experts in the relevant area or studying the relevant law."

Proposed addition (as of December 2024 proposed rulemaking, REG-116610-20):

"Competency includes understanding the benefits and risks associated with relevant technology that is used by the practitioner to provide services to clients or to store and transmit tax return and other confidential information."

The AI Competence Question:

Even without the proposed regulatory amendment, the existing §10.35 competence standard has been interpreted to require technological competence. The OPR has stated that "one facet of competence that has gained importance... is the duty of tax practitioners, support staff, and contractors to maintain sufficient technological competence."

What "competence" means for AI tools:

  1. Understanding what the tool does and how it works: Not at the level of a machine learning engineer, but enough to evaluate its reliability, understand its limitations, and identify when its outputs are likely to be incorrect.

  2. Understanding the tool's limitations: Knowing that LLMs can hallucinate, that they have knowledge cutoffs, that they perform better on some tasks than others, and that they are not infallible legal researchers.

  3. Becoming competent through study: If you are not yet competent with a specific AI tool, §10.35 allows you to become competent "through various methods, such as consulting with experts in the relevant area or studying the relevant law." This means CPE, vendor training, and self-study are legitimate paths to competence, but you must take them before using the tool with client data, not after.

  4. Knowing when to not use AI: A practitioner who lacks competence with an AI tool and cannot develop it must decline to use the tool with client data rather than risk incompetent practice.

⚠️ RISK: Using an AI tool with client data that you do not understand, including not understanding how it handles your data, what it does with prompts, or how reliable its tax outputs are, creates §10.35 exposure. "I use it and it seems to work" is not the same as "I understand the tool's limitations and have verified its performance on the relevant tasks."

§10.36: Supervisory Responsibilities

Full text (key provision):

"(a) Any individual... who has (or individuals who have or share) principal authority and responsibility for overseeing a firm's practice... must take reasonable steps to ensure that the firm has adequate procedures in effect for all members, associates, and employees for purposes of complying with subparts A, B, and C of this part..."

"(b) Any such individual... will be subject to discipline for failing to comply... if, (1) The individual through willfulness, recklessness, or gross incompetence does not take reasonable steps to ensure that the firm has adequate procedures to comply... (2) The individual... does not take reasonable steps to ensure that firm procedures in effect are properly followed... (3) The individual knows or should know that one or more individuals... are... engaged in a pattern or practice... that does not comply with this part... and the individual... fails to take prompt action to correct the noncompliance."

Three Ways §10.36 Liability Attaches:

Failure Type Description AI Application
Failure to Establish Failing to put adequate procedures in place Having no AI usage policy; having no WISP section on AI
Failure to Enforce Procedures exist but are not followed AI policy says "no client data in consumer ChatGPT" but staff use it anyway and you don't monitor
Failure to Correct Knowing noncompliance exists and not addressing it Learning that a staff member has been using an unapproved AI tool and taking no action

For a Solo Practitioner:

If you are the only practitioner in your firm, you have principal authority and responsibility by default. You are both the supervisor and the supervised. This means:

What "Adequate Procedures" Means for AI Use:

An adequate AI use policy under §10.36 would include:

§10.37: Requirements for Written Advice

Section 10.37 governs all written advice on federal tax matters, including electronically transmitted advice. Key requirements:

Application to AI-Generated Written Advice:

When AI assists in drafting a tax opinion, memo, or client advice letter:

  1. The practitioner is responsible for the content. A letter with your firm's name on it is advice from you, not from the AI.
  2. AI-hallucinated citations must be verified. A §10.37 compliant opinion must be grounded in actual law and authorities.
  3. AI-generated assumptions must be reviewed. The standard requires reasonable factual and legal assumptions, assumptions generated by an AI tool may or may not be reasonable.
  4. The "relate law to facts" requirement. AI tools can generate plausible-sounding analysis that applies law to the wrong facts, applies the wrong law to right facts, or conflates multiple clients' facts. The practitioner must ensure the advice actually applies to this specific client's facts.

Documentation for §10.22 Compliance When AI Was Involved

When AI assisted in preparing a return or writing advice, your workpapers should include:

  1. Identification of the AI tool used (vendor, tier/version, date of use)
  2. Description of what the AI was used for (e.g., "Used to extract W-2 data from scanned documents and draft first-draft depreciation schedule")
  3. Evidence of your review: Which AI outputs did you review? How did you verify primary sources? What corrections did you make?
  4. Verification of any cited authority: Note that you confirmed the cited authorities exist and apply as stated
  5. Final judgment applied: A note that professional judgment was applied to all positions taken on the return

📌 PRACTICE TIP: This documentation does not have to be elaborate. A simple workpaper note, "ChatGPT Enterprise used to draft depreciation schedule from client-provided asset list. Schedule reviewed and verified against Form 4562 instructions and §168 authority. Two corrections made: [description]. Final schedule represents my professional judgment.", is far better than no documentation at all and directly supports a §10.22 defense.


Key Takeaways


Quick Review

Q1: An EA signs a federal tax return that contains a deduction the AI tool recommended. The EA reviewed the return but did not verify the AI's cited authority (an IRS Notice that the EA later learns does not exist). Has the EA violated Circular 230?

Answer: Yes, likely §10.22 (due diligence) and §10.35 (competence). Reliance on AI-generated legal analysis without verifying the cited authority does not meet the "reasonable care" standard for relying on another's work product. The EA's professional obligation includes verifying authorities cited in support of tax positions.

Q2: A CPA firm has a written AI use policy stating that no client data may be used with consumer-tier AI tools. A staff accountant routinely uses ChatGPT Plus for client work, and the supervising manager knows about it but takes no action. What is the managing partner's §10.36 exposure?

Answer: The managing partner faces §10.36(b)(3) liability, failure to correct known noncompliance. The manager who knows about the violation and takes no action is subject to discipline. The managing partner may also face §10.36(b)(2) liability for failure to ensure procedures are properly followed.

Q3: A practitioner uses Claude Enterprise to draft a tax opinion letter for a client on a partnership basis question. The practitioner reviews the letter, verifies all cited authorities, makes several corrections to the analysis, and documents the review in the workpapers. Is this compliant with §10.37?

Answer: Likely yes. The practitioner's review, verification of authorities, corrections, and documentation reflect the kind of professional oversight that §10.37 requires. The letter, as issued, represents the practitioner's professional judgment. This is the model workflow.

Q4: A solo CPA practitioner has never reviewed what ChatGPT does with prompt data, has no AI section in her WISP, and has used ChatGPT Free for client data for two tax seasons. She has no client consent and has not analyzed §7216. What Circular 230 sections are potentially at issue?

Answer: Multiple sections: §10.22 (due diligence failure, using a tool without evaluating its data practices or legal implications), §10.35 (competence, using a tool without understanding its benefits and risks), and §10.36 (supervisory responsibilities, even as a solo practitioner, she is responsible for establishing adequate procedures, which she has not done).



← Lesson 7: Revised SSTS (Effective January 1, 2024) · Study Guide home · Lesson 9: State Law Layer →

The AI Lab for Accountants · An educational resource, not legal or tax advice.