Using AI in Tax Practice: The IRS's Six Standards for Doing It Right

Generative AI is becoming a standard part of how tax work gets done — drafting, research, review, client communication. On June 24, 2026, the IRS Office of Professional Responsibility made that official, releasing Issue Number 2026-19: introductory guidelines for the responsible use of generative AI in federal tax practice. Notably, the OPR doesn't discourage AI use. It acknowledges the technology's real benefits for efficiency and fraud detection, and instead answers the question serious practitioners have been asking: what does professional-grade AI use actually look like?

The answer is reassuringly familiar. There are no new rules — the guidance maps AI onto six Circular 230 duties practitioners already follow. Read as a playbook, it comes down to six habits:

1. Review before you rely. Treat every AI output as a first draft. The due-diligence duty means a practitioner reviews AI-created content before it reaches a client file or a filing — the same way you'd review an associate's memo. You own the final word product; AI just gets you there faster.
2. Know your tools. Competence now includes AI literacy. You don't need to understand transformer architecture, but you should know what your tools are good at, where they're weak, and what sources their answers draw from. A tool that shows its reasoning and sources makes this easy; a black box makes it impossible.
3. Let clients share in the efficiency. Circular 230 asks that fees stay reasonable for the work performed, and AI's speed creates genuine cost savings. That's an opportunity: firms that pass some of that time saved through to clients strengthen relationships — and make their AI adoption a selling point.
4. Make it a firm standard, not a personal experiment. Circular 230 expects firms to have procedures ensuring compliant AI use across the practice. Decide which tools are approved, how outputs get verified, and who's responsible — then write it down. A one-page AI policy today saves a scramble later.
5. Verify the authority behind written advice. Written advice must rest on verified factual and legal assumptions. In practice: when an AI answer cites a code section, regulation, or ruling, read the cited authority and confirm it supports the conclusion. This is where tool choice matters most — verification is quick when citations link to real, current sources, and slow when you're hunting to confirm a reference even exists.
6. Keep client data in secure systems. Confidentiality is the bright line. Taxpayer information belongs only in tools secured for it — never in consumer chatbots. Vet vendors, implement data-handling protocols, and train staff on the difference.

Putting it to work

Consider a common scenario in a corporate practice: a client asks how a §382 ownership change affects the use of its net operating loss carryforwards after bringing in new investors. The old workflow is hours in the regulations. The AI-enabled workflow is minutes — ask the question, get a sourced answer, verify the cited authority, and turn it into advice. Every step of that workflow lines up with the OPR's guidance, because the verification is built in rather than skipped.

That's the design principle behind Partnertax.ai's AI Tax Assistant: answers are grounded in a curated corpus of IRS and Treasury sources, and every response carries citations back to the underlying authority. The due-diligence step the OPR asks for — read the source, confirm it says what the answer claims — becomes a click, not a research project.

The bottom line

The OPR guidance is best read as an invitation: AI is welcome in tax practice, on professional terms. Firms that pair capable tools with the six habits above get the efficiency without the exposure — and they'll be ahead of the curve as this introductory guidance matures into settled expectations.

The full bulletin, along with Circular 230 materials, OPR webinars, and FAQs, is available at IRS.gov.

Shashank Sharma is the founder of Partnertax.ai, an AI tax assistant for tax professionals, built on cited IRS and Treasury sources.