Circular 230 + AI: What Tax Pros Can and Can't Do

The IRS regulates tax practitioners through Circular 230. AI doesn't have a carve-out. The rules apply to AI-assisted work the same way they apply to staff-assisted work.

This is not legal advice. Talk to your AICPA professional standards team or counsel for your specific facts.

The rules that matter most

Section 10.22 (Diligence). A practitioner must exercise due diligence in preparing returns and in determining the correctness of any oral or written representations made to the IRS. AI-generated content does not relieve the practitioner of this duty.

Section 10.33 (Best Practices). Tax advisers should communicate clearly with clients regarding terms of engagement, act fairly with the IRS, and establish procedures for compliance with Circular 230. AI tools used in the practice need to fit within these procedures.

Section 10.34 (Standards for Returns). A practitioner may not sign a return knowing it contains a position that lacks a reasonable basis. AI suggesting positions does not change the practitioner's responsibility to evaluate them.

Section 10.35 (Competence). A practitioner must possess the necessary competence to engage in the matter. This includes competence in any tools used. AI users need to be competent in the AI tools' limitations.

What this means in practice

You can use AI for return preparation. Document intake, data entry from PDFs, prior-year comparisons, draft generation. All allowed.

You cannot let AI sign returns. The practitioner who signs is accountable. AI is a tool. Practitioner verifies before signing.

You cannot rely on AI for tax positions without verification. Section 10.34 means you must evaluate any position. AI's confidence is not your confidence.

You should disclose AI use in your engagement letter. Best practice. Not yet a hard requirement under Circular 230, but increasingly expected.

Specific patterns that work

AI for first-draft return prep, practitioner verification before signing. AI extracts data, populates Drake/Lacerte, flags variances. Practitioner reviews everything before signing.

AI for IRS notice triage and draft responses. AI classifies notices and drafts responses. Practitioner reviews and signs. The practitioner remains the responder of record.

AI for tax research orientation, primary source verification by practitioner. AI surfaces relevant code sections and recent rulings. Practitioner verifies on Lexis/CCH/Bloomberg before relying.

AI for client communication drafting. Personalized emails, status updates, reminders. Practitioner reviews and signs out.

Specific patterns that risk violations

Using AI to give tax advice without practitioner verification. Letting AI answer a client's "should I do X?" question without practitioner sign-off.

Auto-filing returns generated by AI without final review. The whole "AI prepares, AI files" workflow without a practitioner step is not Circular 230-compliant for anything more than the simplest scenarios.

Citing AI-generated case law or revenue rulings. AI hallucinates citations. You must verify on a primary source. Practitioners have been sanctioned for filing briefs with fake AI-generated citations. Tax is no different.

Client communication

Most firms now add a paragraph to engagement letters covering AI use. A version that's worked:

"We use AI-assisted tools to support our preparation and review processes. These tools improve efficiency and accuracy of our work. All work product is reviewed by a qualified practitioner before signing. Your data is handled in accordance with our security and confidentiality policies (linked here)."

This is the disclosure pattern. Not legally required everywhere yet. Strongly recommended.

The PII and confidentiality question

Section 7216 of the Internal Revenue Code restricts disclosure of tax return information without client consent. Most AI tools (ChatGPT consumer, Claude.ai consumer) have terms that could be read as third-party disclosure.

Best practice: use enterprise AI tools with no-training and confidentiality terms (Claude for Enterprise, ChatGPT Enterprise, Microsoft Copilot via tenant). Get client consent in your engagement letter if there's any doubt.

The audit-trail question

If the IRS or AICPA peer review asks: how does your firm govern AI use, you need an answer. The answer should include: which tools, what review process, who's accountable, how documentation works.

Firms without this answer are increasingly exposed.

The bottom line

Circular 230 + AI is workable. The practitioner remains accountable. AI is an efficiency tool, not a replacement for professional judgment.

The firms getting in trouble are skipping verification. The firms doing it right are using AI to do MORE per practitioner while keeping the same review discipline.

Not legal advice. Discuss with your firm's compliance team or counsel.