The Office of Professional Responsibility has released new guidance on the responsible use of generative AI in federal tax practice, with an emphasis on one thing:
AI can make tax professionals more productive, but it does not replace professional judgment.
The guidance, published in Issue Number 2026-19, outlines how existing Circular 230 obligations apply when practitioners use generative AI. While the technology is changing rapidly, the professional responsibilities of tax practitioners remain exactly the same.
Here’s what matters most.
The IRS acknowledges that AI has become commonplace across the tax profession. Whether you’re using research platforms, document review software, or modern AI assistants, artificial intelligence is already part of many firms’ workflows.
The IRS also recognizes the significant benefits AI provides:
These benefits come with risks.
Generative AI systems (especially general purpose solutions not built specifically for Tax Professionals) can produce inaccurate information, fabricate citations, introduce bias, or expose confidential client information if used improperly. The IRS makes it clear that practitioners remain responsible for the work they produce.
Every AI-generated document must be reviewed before it reaches a client or the IRS.
That means verifying facts, tax law, calculations, citations, and conclusions.
AI output is a draft, but should not be the final work product.
Competence now includes understanding the strengths and limitations of AI.
Practitioners should understand how their AI tools generate answers, when AI may hallucinate, when additional research is required and how to evaluate AI-generated conclusions.
Written tax advice must continue to consider all relevant facts, apply law to those facts, use reasonable assumptions, and cite appropriate authority.
Blindly copying AI-generated content into client memoranda or IRS correspondence falls short of these standards.
Firm leaders are expected to establish policies governing AI use, this includes staff training, secure handling of taxpayer information, approved AI tools, review procedures, and documentation of AI workflows.
Responsible AI adoption is becoming an operational requirement as much as it is an IT decision.
The IRS specifically reminds practitioners that IRC §§6713 and 7216 continue to apply.
Uploading taxpayer information into unsecured public AI tools could expose firms to civil or criminal penalties.
Choosing secure, enterprise-grade AI systems (like Marble) matters.
The IRS also raises an emerging issue: billing.
If AI significantly reduces the amount of work required, firms should ensure their billing practices accurately reflect the work performed.
Efficiency should benefit both firms and clients.
The most important takeaway is that the IRS is not discouraging AI adoption.
It is encouraging firms to adopt AI responsibly by combining technology with professional judgment.
That means selecting tools that make verification easier, not harder.
The IRS guidance repeatedly emphasizes a simple principle:
AI should assist professional judgment, not replace it.
At Marble, that’s exactly how we’ve designed Marble Intelligence.
The IRS reminds practitioners that they remain responsible for verifying facts, legal authority, and citations before relying on AI-generated content.
Marble provides citation-backed answers that link directly to the underlying federal and state tax authorities, making it easier to validate conclusions rather than relying on unsupported AI output. It’s why we’ve integrated the consolidated tax library into Marble so you don’t need to navigate away in multiple tabs to verify the primary sources.
One of the more practical implications of the IRS guidance is that tax practitioners need more than good answers, they need a well-documented work product.
Research that lives only inside an AI chat window is difficult to review, edit, retrieve later, or incorporate into an engagement file.
That’s why we built Work Products.
With a single click, practitioners can transform a research conversation into an editable document tailored to its intended audience:
These documents aren’t static AI outputs. They’re starting points for professional judgment.
Practitioners must review, edit, and refine the content before saving it as a standalone document within the Client or Project where the research was performed. Every saved Work Product maintains a connection back to the original research conversation, creating clear traceability from research through to the final deliverable.
The IRS also encourages firms to establish documented procedures around AI use.
Rather than leaving valuable research buried inside chat history, Marble stores Work Products within the Client/Project document library where they can be searched, reopened, edited, exported, and referenced in future engagements. This turns AI-assisted research into durable firm knowledge instead of disposable conversations.
The IRS reminds practitioners that confidentiality obligations under IRC §§ 6713 and 7216 remain unchanged.
Marble is built for professional tax firms with enterprise security controls, SOC 2 Type II compliance, and U.S.-hosted infrastructure designed to help firms incorporate AI while protecting sensitive taxpayer information.
Perhaps the biggest takeaway from the IRS guidance is that AI isn’t replacing tax professionals, it’s changing how they work.
The future is conducting research, verifying authority, applying professional judgment, and producing documented work products that can withstand client scrutiny, partner review, and IRS examination.
That’s the workflow Marble is designed to support.
