# Tax Lawyers Must Adapt to AI Now. It’s Already Changed the Game - news.bloombergtax.com

*Источник: news.bloombergtax.com*
*Дата: 2026-07-31*
*Язык: en*

**Кратко:** Artificial intelligence is no longer a future issue for tax lawyers. It is already changing how research is done, how large volumes of documents are reviewed, how contracts and motions are prepared, how audit responses are built, and how clients measure value.

Artificial intelligence is no longer a future issue for tax lawyers. It is already changing how research is done, how large volumes of documents are reviewed, how contracts and motions are prepared, how audit responses are built, and how clients measure value. The profession can argue about whether that change is good or bad. The market wonât care. Tax lawyers have an opportunity to lead the change rather than react to it.
That is the point Spencer Johnson captured in âWho Moved My Cheese,â the 1998 parable about four characters living in a maze and depending on a familiar supply of cheese. When the cheese disappears, two characters rage, wait, and hope the old supply returns. Two others adapt and find new cheese. The lesson is blunt: Change happens; the only real question is whether to move with it or be diminished by it. For tax lawyers, the cheese has moved.
The Anatomy of Resistance
Resistance to change rarely ends because the holdouts are persuaded. It ends because the lawyers who adopt the new tools produce better work, faster and more economically. The market eventually moves around those who donât adapt.
But skepticism shouldnât be confused with irrationality. Many of the concerns lawyers raise about AI are legitimate and deserve careful consideration. For tax lawyers, those concerns typically fall into four categories:
Trust. Not only is that concern fair â it is essential. Unchecked generative AI output shouldnât be trusted, especially on complex tax issues. Current-generation AI can miss nuance, overstate confidence, and produce answers that are incomplete, misleading, or simply wrong. And its most dangerous flaw is the fluency with which it can make those mistakes. It can sound authoritative while misunderstanding the issue, misstating the law, or inventing support that doesnât exist.
Some litigants have learned that lesson the hard way. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), the US District Court for the Southern District of New York sanctioned lawyers who submitted nonexistent judicial opinions with fake quotes and citations generated by ChatGPT. In Park v. Kim, 91 F.4th 610 (2d Cir. 2024), the US Court of Appeals for the Second Circuit referred counsel to its grievance panel after counsel cited a nonexistent state-court decision generated by ChatGPT and admitted she hadnât confirmed its validity. In United States v. Hayes, 763 F. Supp. 3d 1054 (E.D. Cal. 2025), the Eastern District of California sanctioned counsel after a filing cited and quoted a fictitious case the court described as having the markings of a hallucinated AI citation. These are just a few examples.
The tax bar hasnât been immune from this phenomenon. In Clinco v. Commissioner, T.C. Memo. 2026-16, Judge Mark Holmes of the US Tax Court confronted fictitious citations in a brief and reiterated Chief Justice John Robertsâ advice to lawyers who submit briefs containing nonexistent authorities: âAlways a bad idea.â 2023 Year-End Report on the Federal Judiciary 6 (Dec. 31, 2023). Judge Holmes added that "[s]ubmitting a brief with fictitious caselaw is a recipe for sanctions and a clear violation of Rule 11(b) of the Federal Rules of Civil Procedure.â After concluding that several cited authorities were apparent AI hallucinations, he further observed that the attorneyâs argument âcollapses like an overmixed soufflé when one looks at the citations used to prop it up.â
In an earlier case, Thomas v. Commissioner, No. 10795-22, slip op. (order issued Oct. 23, 2024), US Tax Court Judge Ron Buch struck a petitionerâs pretrial memorandum after discovering citations that bore the hallmarks of AI-generated hallucinations. The case names were real, but the cited reporters and docket numbers were incorrect, and the underlying authorities didnât support the propositions for which they were cited. Counsel acknowledged that she hadnât reviewed the memorandum prepared by a paralegal. Given the circumstances, Judge Buch exercised restraint and struck the memorandum without imposing additional sanctions.
Tax Court Rule 33, cited by Judge Buch in Thomas, requires counsel to certify that they have read every submission and that it is well grounded in fact and warranted by existing law. That obligation cannot be delegated to a paralegal, a junior associate, or a language model. The point is especially important when using large language model AI. AI outputs can sound authoritative while being entirely wrong. As Judge Buch explained in Thomas, AI models âhave the ability to give the appearance of understanding text and generating what appear to be thoughtful responsesâ even when they âgenerate[] output that is inaccurate or even nonsensical.â
The lesson isnât subtle: Lawyers canât file the elegant nonsense that AI can create. But that is a lesson about oversight, not abstinence. AI doesnât file briefs; lawyers do. A lawyer who files fake authority generated by AI hasnât been betrayed by technology. The lawyer has simply failed to do the job.
The same goes for a lawyerâs oversight function when an associate misses a case, a search term is too narrow, a database result is misread, or a draft contains a bad cite. Lawyers donât stop using associates, databases, or research platforms because they can produce mistakes. Lawyers supervise, test, verify, and own the final work product. AI requires the same discipline. Bring a healthy amount of skepticism to every output from generative AI.
Quality. Even the leading generative AI platforms now being deployed by large law firms generally produce work product that resembles the output of an average-performing junior associate. The comparison is imperfect, but useful. Like a first-year associate, the technology can often identify relevant issues, organize information, generate a preliminary structure, and produce a draft that moves the project forward. It typically cannot exercise seasoned judgment, distinguish subtle but consequential legal nuances, or reliably determine what matters most to the client.
Yet that limitation shouldnât obscure its value. A draft that would take a junior attorney days to prepare can often be generated in minutes. The fact that the output requires review, revision, and supervision doesnât argue against using the technology; it simply defines the attorneyâs role in the workflow. The generated draft is almost never ready to send to a client or file with a court, but it often meaningfully advances the project.
Generative AI is also an unusually powerful brainstorming tool. It can generate alternative arguments, analogies, examples, planning strategies, and drafting approaches at a pace that few humans can match. For tax practitioners, that ability can be particularly valuable when exploring planning alternatives, identifying factual questions that should be investigated, developing litigation themes, or testing different ways to frame a technical issue.
The challenge is that the technology is often much better at generating ideas than evaluating them or screening out the bad ones. It can produce 10 plausible approaches without reliably identifying which two deserve serious consideration.
Experienced attorneys face a similar dynamic in managing junior lawyers. A productive associate may bring forward numerous ideas, only some of which ultimately survive scrutiny. The difference is that trained attorneys generally develop judgment alongside creativity. They learn to recognize weak arguments, impractical positions, and issues that are unlikely to persuade an auditor or court. Current AI systems are considerably less effective at that filtering function.
For that reason, lawyers should view generative AI as an amplifier rather than a substitute for professional judgment. Its greatest strength isnât that it consistently produces the right answer but that it can rapidly expand the universe of possibilities, generate useful first drafts, and accelerate the development of legal work product. The attorney remains responsible for narrowing those possibilities, rejecting the weak ideas, refining the promising ones, and ultimately exercising the judgment that clients hire lawyers to provide. With the right expectations, the technology can materially increase productivity without diminishing the central role of legal expertise.
(Consistent with the foregoing discussion, I used the two leading generative AI platforms that my firm is currently piloting to edit this article. The technologies generated numerous suggestions; the better ones were incorporated and substantially improved the final product.)
Replacement. âAI is going to take my job.â In some cases, that fear isnât entirely wrong. AI will eliminate some tasks and reduce the need for some billable work. It will make certain forms of research, document review, cite checking, chronology building, issue spotting, and first-draft generation faster, cheaper, and less dependent on large numbers of human hours. That isnât a reason to deny the technology. It is the point.
Clients buy judgment, strategy, risk reduction, and results. If AI compresses the mechanical work, the lawyer should spend more time on the judgment work. Tax planning isnât mere information retrieval. Neither is defending a position before the IRS. The hardest work isnât finding the rule; it is knowing how the rule will operate in the clientâs real world â across business objectives, cash-flow constraints, audit risk, accounting treatment, investor expectations, and reputational concerns. AI can surface authorities, organize facts, compare structures, and pressure-test arguments. But it is less effective at discerning when a technically available position is too aggressive, when a clean structure is better than a clever one, when a client needs a practical answer instead of a law-review answer, or when restraint is the strongest advocacy move. AI can accelerate the work around judgment. It canât replace judgment.
Every generation of lawyers has faced a version of this shift. Photocopiers replaced carbon paper. Fax machines replaced overnight mail. Personal computers, word processors, and spreadsheet applications replaced countless hours of manual drafting and calculation. Online legal research databases reduced the need to search reporters and digests by hand. Email, document-management systems, and e-discovery platforms transformed communication, file organization, and document review. None of these innovations eliminated lawyering. They simply reduced the time spent on tasks that clients ultimately stopped wanting to purchase as separate legal work.
Generative AI is the next version of the same story, only faster and more powerful. The work that survives is the work that requires judgment: deciding what matters, what is missing, what is risky, what is persuasive, what is privileged, what is strategically useful, and what should never leave the building. And it includes oversight over AI-generated output, which, under the current generation of AI products, is typically first-year associate quality work at best.
A related anxiety is about identity. Tax lawyers invested years developing skills that enabled them to perform many of the tasks AI can now assist with, and they paid their dues through countless hours doing that work. It is natural to view those tasks as part of what makes us lawyers. But that is nostalgia mistaken for principle. The essence of lawyering has never been performing every mechanical step by hand. It is exercising judgment, accepting responsibility for the result, and performing the countless functions that clients look to trusted advisers to provide.
Guardrails. AI is powerful, and powerful tools need guardrails. Email can waive privilege. Search engines can return bad sources. E-discovery tools can miss documents.
The answer isnât to reject these tools; it is to understand the risks, build the guardrails, and use the tools responsibly. AI in tax practice should be approached the same way: neither with blind trust nor fear, but with disciplined, lawyer-led adoption.
What Competence Requires
Proficiency with generative AI tools is rapidly becoming a core competency for tax lawyers. It is also consistent with a tax lawyerâs ethical obligations.
Professional competence includes technological competence. Model Rule of Professional Conduct 1.1 requires lawyers to keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology. That reflects a basic professional reality: Lawyers canât make informed choices about tools they refuse to understand.
This doesnât mean adopting every AI product, putting confidential client information into an unapproved open-system tool, or outsourcing judgment to a machine. It means knowing enough to decide when AI can improve the work, when it shouldnât be used, what verification is required, what confidentiality limits apply, and what human review remains essential.
That understanding takes practice. Lawyers need to learn which prompts work, which tools are reliable for which tasks, where hallucinations tend to appear, how to demand citations, how to test those citations, how to preserve privilege and confidentiality, and how to document the review process. That is what lawyers do when the practice changes.
Responsible AI use is consistent with existing Circular 230 requirements. Section 10.22 (31 C.F.R. § 10.22) requires due diligence in verifying facts, citations, and calculations, and Section 10.36 (31 C.F.R. § 10.36) requires firms to maintain reasonable supervisory functions. Those obligations apply whether the work originates from a lawyer, a paralegal, or an AI tool.
The key insight across all of these frameworks is consistent: The problem isnât AI use, but unverified AI use. As multiple courts have emphasized, a fake citation is no more acceptable from an AI tool than from a careless associate. The standard of care hasnât changed.
The Disclosure Moment
AI use also raises a separate and increasingly important question: When does it need to be disclosed? The answer isnât one-size-fits-all, but the direction of travel is clear. Lawyers must know the applicable court rules, standing orders, judge-specific procedures, client requirements, and engagement obligations before using AI in a matter.
If a court requires disclosure of AI use, disclose it. If a judge requires certification that authorities have been independently verified, make the certification only after the work has actually been checked. If a client has restricted AI use, follow the restriction. If confidential information would be entered into a tool that isnât approved for that use, donât enter it.
The profession is nearing an inflection point. Lawyers shouldnât be embarrassed to disclose AI use. If anything, the greater question may eventually be why a lawyer chose not to use tools capable of improving the work. The issue isnât whether AI was used, but whether it was used with enough discipline to improve the work product. Responsible AI use is a way to do lawyering better.
Key Takeaways
AI isnât the end of tax practice. It is a force multiplier for lawyers willing to learn how to use it. It shortens the distance between raw information and legal judgment. It lets lawyers spend less time hunting for the cheese and more time deciding what to do once they find it.
The cheese isnât coming back. The maze has changed. The question is no longer whether tax lawyers should go looking for new cheese, but how fast they are willing to move â and whether they will help lead the profession through the maze.
The author used AI to draft a portion of this article.
This article does not necessarily reflect the opinion of Bloomberg Industry Group Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.
Author Information
Peter Lowy is a partner at Nelson Mullins.
Interested in writing? Review our author guidelines and submit pitches to Insights@bloombergindustry.com.
Learn more about Bloomberg Tax or Log In to keep reading:
See Breaking News in Context
From research to software to news, find what you need to stay ahead.
Already a subscriber?
Log in to keep reading or access research tools and resources.

[Оригинал](https://news.bloombergtax.com/daily-tax-report-international/tax-lawyers-must-adapt-to-ai-now-its-already-changed-the-game)