Tax & AccountingJuly 02, 2026

The IRS just drew a clearer line on AI in tax practice. Here's what it means for your firm

Key Takeaways

  • The IRS OPR's guidance doesn't rewrite Circular 230. It restates and confirms that AI-assisted work is still the practitioner's work.
  • Due diligence, competence, written advice, confidentiality, and firm procedures under Circular 230 all still apply, and Section 7216 data protections remain a related concern for AI use.
  • Firms should be thinking about how AI changes billing conversations with clients, not just workflow.
  • The quality and trustworthiness of the content behind an AI tool matter as much as the tool itself.

There was a lot of speculation about what the IRS would say when it finally weighed in on generative AI in tax practice. Would it crack down? Slow adoption? Introduce a new compliance regime?

The answer, as it turns out, is none of the above. The Office of Professional Responsibility's introductory guidance doesn't reinvent anything. It takes Circular 230, which has governed practitioner conduct for decades, and confirms that it applies to AI-assisted work, too.

That's the whole thesis. And honestly, it should be.

What the guidance actually says

The OPR connected AI use to several familiar Circular 230 obligations that apply to any work product a practitioner puts their name on: Section 10.22, Section 10.35, Section 10.36, and Section 10.37

The guidance also raises data-handling concerns that naturally connect to Section 7216, the statute that governs the disclosure and use of tax return data. The OPR doesn't cite 7216 by name, but the underlying concern is the same. When taxpayer data is fed into a generative AI tool, the firm needs to be able to answer a straightforward question: where did that data go, and who could see it? 

Alongside those obligations, the guidance names the risks that have shaped every serious conversation about generative AI. Hallucinations that read as confident but cite cases that don't exist. Bias that's hard to see. Confidentiality exposure when taxpayer information passes through systems the firm doesn't control. And billing questions when AI compresses the hours a client used to be charged for.

The through-line is straightforward: AI-generated text is a draft, not a deliverable. A practitioner reviews, verifies, and takes responsibility for it.

The signature at the bottom of a return still means something, and it still means the same thing it did before ChatGPT existed.

The billing question worth sitting with

The guidance raises a business question that deserves more attention than it's getting: what happens to the billing model when AI compresses research and drafting time?

For firms that bill hourly, the answer isn't obvious. Passing along the full traditional rate for work an AI tool produced in a fraction of the time creates its own ethical exposure. Discounting it may make sense for the client, but it raises hard questions about how firms recover the investment they're making in AI tools and the professional oversight those tools still require.

Firms should be thinking about this now, not after clients start asking. What does the engagement letter say? How are AI-assisted deliverables described? Where does the value of the practitioner's judgment show up in the invoice? The answers will shape how much room firms have to invest in the AI tools that actually hold up under professional scrutiny.

Content quality matters as much as the model

There's a natural instinct to evaluate AI tools by looking at the model itself. How large is it? How current? How fast? Those are reasonable questions, but they miss where most of the risk actually sits.

The quality of an AI tool in tax practice comes from the content underneath it. A general-purpose model trained on the open internet doesn't know the difference between a controlling authority and a blog post. It doesn't know which regulations have been superseded, or that a court's reasoning was later distinguished or overturned. That's not a knock on the technology.

It's a description of what it can and can't do without expert-validated content behind it.

Firms relying on AI tools grounded in trusted tax and regulatory content, refreshed by subject-matter experts, and traceable back to authoritative sources, are in a different risk category than firms relying on general-purpose AI to answer tax questions. The OPR guidance doesn't say that in so many words, but the responsibilities it reinforces make the distinction plain.

Where Wolters Kluwer's approach fits

Wolters Kluwer's Expert AI capabilities inside CCH AxcessTM are built on a specific point of view: AI belongs inside the professional's workflow, grounded in trusted content, with the practitioner making the final call. Embedded in the platform firms already use, with admin controls, role-based access, and enterprise-grade infrastructure behind it.

What "grounded in trusted content" means in practice is the part that matters most for the Circular 230 conversation. The AI draws on the same tax and regulatory content that Wolters Kluwer's editorial teams maintain, cite-check, and update. When a practitioner reviews an AI-surfaced insight, the underlying sources are traceable. That's the difference between a black-box answer and one a professional can stand behind.

The design also reflects the practical reality of the data-handling and firm governance obligations tax practitioners already work under, including Section 7216. Taxpayer data stays within the platform's boundaries rather than being routed through general-purpose AI services. Firm administrators can define who has access to which tools, on which client data, with an audit trail behind it. None of that eliminates the practitioner's responsibility, but it gives the firm a defensible foundation to build on. Governance, source confidence, and data handling should be part of the design from day one. Adding them later never works as well as intended.

What firm leaders should be thinking about today

  1. Start with guardrails, not a perfect framework. Document who can use what tools, on what data, and how outputs get reviewed.
  2. Treat AI outputs as drafts. Always. Build review into the workflow.
  3. Look at the content, not just the model. Trusted, expert-validated content is where AI in tax practice earns credibility.
  4. Rethink the billing conversation. Get ahead of what AI-assisted work means for engagement letters and invoices.
  5. Recognize this isn't just a tech decision. It's a decision on governance, data protection, and professional responsibility.

AI is here, and it's welcome. It just comes with the same responsibilities that have always defined tax practice, and firms should be watching closely as the profession figures out how to meet them.

headshot of a woman with dark curly hair
Vice President, Tax Compliance Solutions for Wolters Kluwer Tax & Accounting North America
Back To Top