On June 24, 2026, the IRS Office of Professional Responsibility issued Alert 2026-19, Introductory Guidelines for Responsible AI Use in Federal Tax Practice. It is the first time the OPR, which enforces Treasury Circular 230 and oversees conduct of CPAs, attorneys, enrolled agents, enrolled actuaries, and other practitioners before the IRS, has formally mapped how existing professional obligations apply to the use of artificial intelligence.
The alert does not create new rules. It applies existing ones to a new fact pattern. Every Circular 230 obligation you currently carry still runs when AI produces part of your work. Due diligence, competence, fee standards, written advice requirements, client confidentiality, and firm supervisory procedures each apply to AI-generated content with the same force they apply to content you wrote yourself.
What changed on June 24 is that the OPR removed the excuse of uncertainty. A practitioner who relied on AI-generated content before this alert could theoretically argue they were unsure whether their existing professional duties extended to AI outputs. After Alert 2026-19, that argument is gone. The OPR has clarified explicitly, by name, how each relevant Circular 230 section applies to AI. Circular 230 cases follow from that clarity.
This post covers each of the six Circular 230 provisions the OPR addressed, the specific compliance implications of each, and the documentation framework the OPR expects firms to have in place now.
What Is IRS OPR Alert 2026-19 and Why Is It Different From Prior AI Guidance?
The IRS Office of Professional Responsibility enforces Treasury Department Circular 230, 31 C.F.R. Part 10. Circular 230 sets the standards of practice for all practitioners who represent taxpayers before the IRS, including attorneys, CPAs, enrolled agents, enrolled actuaries, and other persons. The OPR can investigate complaints, issue censures, suspend or disbar practitioners, and refer matters to the Department of Justice.
Prior to June 24, 2026, the OPR had issued no formal guidance on AI use in tax practice. Various bar associations, state CPA societies, and accounting firm internal guidance documents had addressed AI risk from their own professional perspectives, but the IRS itself had not spoken.
Alert 2026-19 is titled "Introductory Guidelines for Responsible AI Use in Federal Tax Practice." The word introductory is important and has been misread by some practitioners as a signal that the guidance is tentative or non-binding. That reading is incorrect. The TLY analysis confirms the correct interpretation: the guidance is introductory because it applies only to the current state of AI technology. The OPR expects to issue additional guidance as AI tools and regulatory frameworks evolve. Introductory means there is more to come, not that the current guidance is optional.
The OPR defines AI broadly as the use of machines in a way that mimics human cognitive skills, including judgment, perception, and prioritization. The guidance distinguishes legacy AI tools (advanced legal research platforms, document review tools, established analytics software) from generative AI, which it defines as systems that can make discretionary decisions with minimal or no human interaction, producing original content including documents, analysis, research, and tax advice.
The alert's coverage is focused on generative AI and the specific risks it creates: fabricated citations, invented facts, hallucinated authority, and the appearance of authoritative accuracy in outputs that are not factually correct. The OPR references the real-world consequences of AI hallucinations in practice. It specifically cites the case of a 230-page report that Deloitte Australia produced for the Australian government in July 2025 that contained fabricated quotes attributed to a judge, references to sources that did not exist, and books attributed to the wrong author. The government published the report before the errors were discovered. For tax practitioners, the same category of error carries the same risk of harm to clients and the same regulatory exposure.
Obligation #1: Due Diligence Under Circular 230 Section 10.22, What AI Cannot Do for You
Section 10.22 of Circular 230 requires practitioners to exercise due diligence in the preparation and filing of tax returns and documents relating to IRS matters, and in oral or written representations to the IRS or to clients.
The OPR's application to AI is direct. Practitioners must thoroughly review all AI-created documents and language incorporated into writings before delivery to a client or submission to the IRS. The federal standard of due diligence cannot be delegated to an algorithmic process. The duty requires active verification of the accuracy of facts, citations, and calculations produced by AI. Human scrutiny and editing are essential to ensure correctness and compliance with IRS expectations.
The specific due diligence obligations the OPR maps to AI use:
Citation verification. When AI produces a legal citation, whether to a tax code provision, a Treasury regulation, a Revenue Ruling, or case law, the practitioner must independently verify that the cited authority exists, says what the AI claims it says, and has not been superseded, overruled, or modified. The nexairi analysis is specific: if AI cited a 2018 ruling, confirm it has not been superseded. A confident, well-formatted citation is not the same as a correct one.
Calculation verification. When AI produces a tax calculation, estimated tax amount, or financial analysis, the practitioner must verify the calculation against the source numbers. AI arithmetic errors and formula errors occur and are not always obvious in the output.
Factual accuracy. When AI drafts a statement of facts about a client's situation (for a tax opinion, a tax return position, or a representation to the IRS), the practitioner must verify that the stated facts are accurate and that all material facts have been included, not just those that support the AI-generated position.
The due diligence obligation means that the AI output is the starting point for professional work, not the finished product. The practitioner who signs the return, opinion, or filing is the person responsible for everything in it, regardless of how it was produced.
The most important documentation implication: the OPR will expect evidence of the due diligence review when it investigates a complaint. A notation in the engagement file that says "reviewed by [initials]" is not adequate. The nexairi analysis confirms the OPR's expectation explicitly: document who reviewed the output, what the review consisted of, and how the reviewer confirmed adequacy. The documentation standard for AI review is the same as the documentation standard for any other professional work product.
Obligation #2: Competence Under Circular 230 Section 10.35, Why Technological Literacy Is Now Part of Professional Competence
Section 10.35 of Circular 230 requires that practitioners possess the necessary competence to practice before the IRS. Competent practice requires the appropriate level of knowledge, skill, thoroughness, and preparation necessary for the matter.
The OPR's alert expands the definition of competence to include technological literacy in the context of AI use. Competence now requires an understanding of both the law and the technology used in client representation, including AI systems' operational mechanics, limitations, and risks. A practitioner must possess the capability to understand how AI develops content, recognise the potential for bias or errors, and be able to evaluate whether AI outputs are suitable for use in IRS matters.
Three specific components of AI competence the alert identifies:
Understanding how AI generates content. Practitioners must understand, at a working level, the mechanisms by which their AI tools produce outputs. For generative AI, this includes understanding that the system predicts likely text sequences rather than retrieving factually verified information, that it has a training data cutoff date that may not reflect current law, and that its confidence in its output does not correlate with the output's accuracy.
Recognising potential for bias and errors. AI training data reflects the biases in that data, and generative AI systems can reproduce and amplify those biases in their outputs. Tax practitioners must be able to identify when AI output appears to reflect training data patterns rather than the actual law applicable to the client's specific facts.
Evaluating when AI output is unsuitable. Some tasks are inappropriate for AI use in tax practice regardless of the practitioner's level of review. AI outputs that depend on opaque or unexplainable reasoning, that involve novel legal questions not well-represented in training data, or that require judgment calls about client-specific facts that AI has not been given are examples of outputs where the practitioner must recognise AI limitations and work from primary sources.
The competence standard applies to continuing education. The OPR expects practitioners to stay current as AI tools and regulatory guidance evolve, consistent with the continuing professional education obligations applicable to their licensing.
Obligation #3: Fees Under Circular 230 Section 10.27(a), The Billing Problem Nobody Wants to Talk About
Section 10.27(a) of Circular 230 provides that a practitioner may not charge an unconscionable fee in connection with any matter before the IRS. The OPR's alert applies this provision to AI use in a way that has significant billing practice implications.
The logic is straightforward. If AI tools reduce the time needed for research and drafting, and a practitioner bills the client for the full manual-labour time that was not actually spent, the fee may not accurately reflect the value of the services provided and could constitute an unconscionable fee. The OPR's alert makes explicit what many practitioners had been choosing not to examine: cost savings from AI must be passed on to clients, and billing practices must openly reflect the efficiencies gained.
Three specific billing issues the OPR flags:
Double-billing. Charging for AI-assisted work as if it were performed manually, when AI materially reduced the effort required, is explicitly identified as a potential Section 10.27(a) violation. The Thomson Reuters analysis describes this as a pattern the OPR warns could trigger enforcement.
Lack of disclosure. Practitioners should disclose the AI activities performed in the engagement and fairly credit any cost reductions to the client's account. The disclosure does not need to be elaborate, but it should acknowledge that AI tools were used and that the fee reflects the efficiency gained.
Fee differential patterns. A noticeable pattern of billing differentials, where certain tasks are systematically billed at a rate inconsistent with the time actually spent because of AI assistance, may constitute a violation. The OPR specifically says a noticeable pattern of billing differentials may constitute a Section 10.27(a) violation.
The practical implication for firms: pull a recent sample of engagements where generative AI was used in research or drafting. Compare the time billed to the time that would have been required without AI. If there is a material difference that was not reflected in the fee or disclosed to the client, the firm has a current billing exposure that should be addressed prospectively and, where the fee impact was material, considered retroactively.
The billing issue is also an opportunity. Firms that proactively disclose AI use and transparently communicate how AI efficiency benefits clients will be better positioned in competitive pitches than firms that continue billing at traditional rates without addressing how AI is affecting their service delivery.
Obligation #4: Firm Supervisory Procedures Under Circular 230 Section 10.36, What Firm Leaders Are Now Personally Responsible For
Section 10.36 of Circular 230 requires that principals and employees who have supervisory authority over the firm's practice take reasonable steps to ensure that the firm has adequate procedures in effect for purposes of complying with Circular 230, and to ensure compliance with those procedures.
The OPR's alert makes Section 10.36 a firm-level AI governance obligation. Firm leaders, partners, managing directors, and others in supervisory positions are now personally responsible for ensuring the firm has implemented specific AI governance policies and procedures. The OPR enumerates three specific firm obligations:
Training. The firm must provide comprehensive training to all staff on the risks and requirements of using AI in tax practice, specifically covering the Circular 230 obligations addressed in Alert 2026-19. Staff who use AI tools without understanding those obligations expose the firm to vicarious liability under Section 10.36.
Documented protocols. The firm must establish and document protocols for secure data handling, monitoring AI accuracy, and reviewing AI-generated content before it is used in client work or submitted to the IRS. The documentation requirement is explicit: the OPR expects to see documentation of the procedures when it reviews a firm's practices.
Third-party AI tool vetting. The firm must thoroughly vet any third-party or outsourced AI tools before allowing them to be used in practice. The vetting should address: Does the tool retain client data? Where is data stored? Who has access to it? What is the vendor's security certification? Has the tool been tested for reliability in the specific tax tasks for which it will be used?
The supervisory obligation means firm leaders cannot claim ignorance if staff use consumer AI tools without oversight. A partner who knows that associates are using public generative AI platforms to draft client correspondence but has not implemented a policy addressing this has failed the Section 10.36 obligation regardless of whether any specific harm has resulted.
The Foster Garvey analysis confirms the personal liability dimension: the obligations under Section 10.36 fall on the people who run the practice. A firm-level failure to implement adequate AI procedures is not just an institutional compliance issue; it is a personal disciplinary risk for each person with supervisory authority.
Obligation #5: Written Advice Under Circular 230 Section 10.37, The Opaque Logic Problem
Section 10.37 of Circular 230 addresses written tax advice, which includes written opinions on the tax consequences of transactions, written advice in correspondence, and written advice incorporated into transactional documents. The section requires that written advice be based on reasonable factual and legal assumptions, that the practitioner use reasonable efforts to identify relevant facts, and that the practitioner not rely on unreasonable representations.
The OPR's application to AI presents a specific problem for practitioners who rely on generative AI output in written advice: if the AI system's underlying reasoning is opaque, meaning the practitioner cannot trace the system's analysis to the specific authority it is relying on and verify that the authority is accurate and applicable, relying on that output may itself constitute unreasonable reliance.
The KTYSONLAW analysis states the principle directly: "Blind reliance on AI yields may constitute unreasonable reliance." The OPR alert frames this as a risk when an AI system's underlying logic or services are unclear.
The practical compliance requirement for written advice:
Every factual assumption on which the written advice relies must be independently verified as accurate, even if AI drafted the statement of facts.
Every legal citation on which the written advice relies must be independently verified as existing, as saying what the advice claims it says, and as applicable to the client's specific facts.
Every projection or financial forecast included in the written advice must be independently verified against source data.
The written advice obligation extends to advice provided by email, in letters, and in any other written medium. It is not limited to formal written opinions. A practitioner who uses AI to draft a client email containing advice on a tax position is subject to Section 10.37 for that email.
The reasonable assumptions requirement has always existed. Alert 2026-19 makes clear that the practitioner cannot satisfy that requirement by pointing to the AI system's output as the basis for the assumptions. The assumptions must be verified by the practitioner independently.
Obligation #6: Confidentiality Under IRC Sections 6713 and 7216, Why Client Data and Public AI Platforms Cannot Mix
Circular 230 Section 10.51(a)(15) identifies willful disclosure of tax return information without client consent as incompetent and disreputable conduct subject to disciplinary action. IRC Section 6713 imposes civil penalties for unauthorised disclosure of tax return information. IRC Section 7216(a) imposes criminal penalties for wilful unauthorised disclosure.
The OPR's alert addresses this obligation directly in the AI context. The specific risk it identifies: practitioners who upload client tax return information, social security numbers, financial data, or other sensitive taxpayer information to unsecured or public AI platforms are potentially violating these provisions, because the data may be retained by the AI system's operator, processed by the system in ways that expose it to other users, or stored in databases that are not under the practitioner's control.
The OPR's instruction is unambiguous: practitioners must strictly handle all client data using only secure, enterprise-approved AI systems with robust confidentiality safeguards in place.
The three-part test a firm should apply before using any AI tool with client data:
Enterprise approval. Has the firm's IT and compliance leadership reviewed and approved this specific AI tool for use with client data? Consumer-grade AI products with public processing pipelines are not enterprise-approved absent a specific contractual arrangement with the vendor that addresses data retention and confidentiality.
Confidentiality safeguards. Does the AI vendor have a BAA-equivalent contractual commitment that the firm's client data will not be retained, will not be used to train the AI model, and will not be accessible to the vendor's staff or other users?
Documented retention policy. Does the firm have documentation of the tool's approval, the vendor's data handling commitments, and the procedures for using the tool with client data?
The confidentiality obligation applies to every member of the firm's staff, not just licensed practitioners. A tax associate who uploads a client's prior-year return to a public AI platform to help draft research memos has potentially triggered IRC Section 6713 liability regardless of whether any disclosure outside the AI platform actually occurred.
The Deloitte Australia Warning: What AI Hallucinations Look Like When They Hit Client Work
The OPR's alert specifically references the Deloitte Australia incident from July 2025 as a real-world example of the professional consequences of inadequate AI output review. The report produced for the Australian government contained fabricated quotes attributed to a judge, references to sources that did not exist, and books attributed to the wrong author. The government published the report before the errors were discovered. The OPR reference is not incidental: it signals that the tax practice regulatory community is watching the AI hallucination problem in adjacent professions and expects tax practitioners to learn from those examples.
The specific categories of AI hallucination that OPR Alert 2026-19 implicitly addresses through its citation verification and due diligence requirements:
Fabricated case citations. AI systems frequently generate plausible-sounding but non-existent case citations. A case name, docket number, and court citation can appear authoritative while corresponding to no actual case. Any tax opinion or research memo that relies on case citations must independently verify each citation against Westlaw, LexisNexis, or the official court databases.
Invented Revenue Rulings or Regulations. AI systems can generate Revenue Ruling numbers, regulation references, and other IRS guidance citations that do not exist. A Revenue Ruling cited in a tax advice letter must be verified against the Internal Revenue Bulletin and the official IRS database.
Superseded or modified authority. Even where an AI citation refers to a real source, the source may have been superseded, overruled, or modified since the AI's training data cutoff. A 2021 Tax Court case that was reversed on appeal in 2024 may appear in AI output without the qualification that it was reversed.
Incorrect factual representations. AI systems summarising a client's tax situation may include incorrect facts, misapply client-specific information, or omit material facts. The practitioner must verify that the stated facts in any written work product are accurate and complete.
The professional consequences the courts have imposed for AI hallucination failures in the legal profession, referenced in the OPR alert, include significant financial sanctions, public censure, and mandatory ethics courses. The OPR's alert makes clear that tax practitioners face the same risk under Circular 230.
What Firm Leaders Must Do Before the Next Filing Season: A Six-Step Action Plan
Six specific actions that satisfy the OPR Alert 2026-19 implementation obligations.
Step 1: inventory every AI tool currently in use. Survey all staff to identify which AI tools are being used, for what purposes, and with what client data. The inventory should cover research AI, drafting AI, document review AI, and any AI-integrated features in the firm's existing software (tax preparation software, document management, time and billing). This inventory is the starting point for Steps 2 through 5.
Step 2: establish the approved tools list. Review each AI tool in the inventory against the confidentiality obligation criteria described above: enterprise approval, confidentiality safeguards, and documented retention policy. Remove from active use any tool that does not satisfy all three criteria for use with client data. Maintain a list of approved tools with the date of approval and the approval basis.
Step 3: build the due diligence documentation protocol. For each type of AI-assisted work product (research memo, tax return position, client letter, written advice), define what an adequate review of the AI output looks like. Document the review standard. Implement the documentation practice in the engagement file for every AI-assisted work product.
Step 4: train all staff on Alert 2026-19 requirements. All professional staff who use or may use AI tools must receive training that covers the six Circular 230 obligations described above, the firm's approved tools list, the due diligence review protocol, and the confidentiality requirements. Document the training with dates and participants.
Step 5: review recent billing practices for AI efficiency compliance. Pull a sample of recent engagements where AI was used in research or drafting. Assess whether fees were adjusted to reflect AI efficiency savings. For prospective engagements, implement a billing disclosure and adjustment practice. For past engagements where the fee impact was material and was not addressed, consult with the firm's ethics counsel about the appropriate prospective and retroactive steps.
Step 6: establish a continuing education commitment. Alert 2026-19 is the OPR's introductory guidance. More guidance will follow as AI tools and the regulatory framework evolve. Assign responsibility within the firm for monitoring future OPR guidance, AICPA AI guidance, applicable bar guidance, and relevant court decisions. Schedule an annual review of the AI compliance policy.
Frequently Asked Questions
What is IRS OPR Alert 2026-19?
OPR Alert 2026-19, titled "Introductory Guidelines for Responsible AI Use in Federal Tax Practice," was issued by the IRS Office of Professional Responsibility on June 24, 2026. It is the first formal guidance from the OPR on how existing Treasury Circular 230 professional obligations apply to the use of artificial intelligence, particularly generative AI, in tax practice. The alert does not create new rules but clarifies how six existing Circular 230 provisions apply to AI-generated content.
Does Alert 2026-19 apply to all tax practitioners?
Yes. The OPR enforces Circular 230 for all practitioners who practice before the IRS, including attorneys, CPAs, enrolled agents, enrolled actuaries, and others. Alert 2026-19 applies to every practitioner in those categories who uses AI tools in tax work for IRS-related matters.
What does Circular 230 section 10.22 require for AI output?
Section 10.22 requires due diligence in preparing and filing tax returns and documents. The OPR clarifies that this standard applies to AI-generated content. Practitioners must thoroughly review all AI-created documents for accuracy, independently verify all citations and calculations, and not rely solely on AI output without verification. The due diligence obligation cannot be delegated to the AI system.
Can practitioners use generative AI to draft written tax advice?
Yes, but the AI output is a starting point only. Under Section 10.37, written tax advice must be based on reasonable factual and legal assumptions that the practitioner has independently verified. AI-generated citations and factual statements must be verified before the advice is delivered to a client. If the AI system's reasoning is opaque and cannot be traced to verified authority, relying on that reasoning may constitute unreasonable reliance under Section 10.37.
Does AI use affect billing under Circular 230?
Yes. Section 10.27(a) prohibits unconscionable fees. When AI reduces the time needed for research or drafting, billing for the full manual-labour time that was not spent may violate this provision. The OPR expects cost savings from AI to be reflected in client billing and recommends disclosure of AI use. Double-billing for AI-assisted work and then charging as if the work were done manually is specifically identified as a potential violation.
What are the confidentiality risks of using AI with client data?
IRC Sections 6713 and 7216(a) impose civil and criminal penalties for unauthorised disclosure of tax return information. Uploading client data to unsecured or public AI platforms creates risk of unauthorised disclosure if the platform retains client data, uses it for model training, or exposes it to other users. The OPR requires that client data be handled using only secure, enterprise-approved AI systems with documented confidentiality safeguards.
Is Alert 2026-19 binding, and what happens if practitioners violate it?
Alert 2026-19 clarifies existing Circular 230 obligations. Those obligations were already binding. A practitioner who violates them through AI-related conduct faces the same Circular 230 disciplinary consequences as any other violation: reprimands, censures, suspensions, disbarments, or referrals to the Department of Justice for criminal violations. The guidance cannot be dismissed as non-binding because it is labelled introductory; the underlying Circular 230 sections were always enforceable.
Key Takeaways
- On June 24, 2026, the IRS Office of Professional Responsibility issued Alert 2026-19, Introductory Guidelines for Responsible AI Use in Federal Tax Practice. It is the first formal OPR guidance mapping Circular 230 to AI use in tax practice.
- The alert does not create new rules. It applies six existing Circular 230 provisions to the new fact pattern of AI-generated content. Practitioners who used AI before the alert should reassess their practices against these six provisions.
- Section 10.22 (Due Diligence): practitioners must independently verify all AI-generated facts, citations, and calculations before submitting to the IRS or delivering to clients. Documentation of the review is expected.
- Section 10.27(a) (Fees): cost savings from AI must be reflected in client billing. Billing for time that AI saved without adjusting the fee may constitute an unconscionable fee. Disclosure of AI use is expected.
- Section 10.35 (Competence): technological literacy is now part of professional competence. Practitioners must understand how their AI tools work, where they fail, and when AI output is unsuitable for IRS use.
- Section 10.36 (Firm Procedures): firm leaders are personally responsible for implementing AI training, data security protocols, output monitoring, and third-party tool vetting. A firm-level AI policy is required, not optional.
- Section 10.37 (Written Advice): AI-generated written advice must be based on reasonable assumptions that the practitioner has independently verified. Opaque AI reasoning that cannot be traced to verified authority may constitute unreasonable reliance.
- IRC Sections 6713 and 7216 (Confidentiality): client data cannot be uploaded to unsecured or public AI platforms. Only enterprise-approved AI systems with documented confidentiality safeguards may be used with client tax information.
- The Deloitte Australia incident (July 2025), cited in the alert, involving fabricated citations and non-existent sources in a government-published report, illustrates the professional consequences the OPR expects practitioners to learn from and avoid.
