Microsoft Copilot AI Hallucinates on Chevron Deference (4/12/26)

I have previously written on artificial intelligence (“AI”) related to tax procedure issues. I write today to address AI hallucinations of the type that have been incorporated into briefs and, more rarely, even court opinions. By hallucinations, I mean (from Wikipedia, here): “a response generated by AI that contains false or misleading information presented as fact.” The specific instance I address here is false readings of judicial opinions in the context of Chevron deference.

Readers will recall that I have previously discussed the issue of Chevron deference and the demise of Chevron deference in Loper Bright. One of the issues I addressed but return to today is whether courts determined the best interpretation was an interpretation other than an agency interpretation but nevertheless deferred under Chevron to the agency not-best interpretation. My claim was that Chevron required statutory ambiguity which meant that the court could not determine the best interpretation; only when ambiguity existed could the court defer to (apply) the agency interpretation. Stated otherwise, if a court could determine the best interpretation, the statute was not ambiguous and the best interpretation applied at Chevron Step One (i.e., no deference to the agency interpretation). Extended, the claim is that courts did not defer to an agency interpretation when it was not the best interpretation; rather, courts only deferred when they could not determine the best interpretation. At least that is how Chevron should have worked.

Yesterday, I returned to AI, specifically MS Copilot, to address this issue. I link here for download an pdf from an MS Word file with my prompts (4 prompts with prompts 2-4 following through on the first prompt). I provide here prompts 1-4 which refine the inquiry and provide the Copilot response only to the final prompt:

PROMPT #1: can you please identify cases decided under Chevron in which a court explicitly stated its best interpretation and said that it was nevertheless deferring to a not-best agency interpretation?

PROMPT #2 Please provide direct quotations from each case.

PROMPT #3: Please provide pinpoint citations for each case. [JAT Note: pinpoint citations are local page cites.]

PROMPT #4: Please provide cases and quotes from cases where judges criticized Chevron for forcing them to accept worse interpretations.

I will copy and paste the answer to Prompt #4, but first a big red flag was presented in the response to PROMPT #2 when I asked for direct quotes. The response started with the following:

I can absolutely give you direct quotations, but there’s one important constraint: I can only quote one or two lines verbatim from each case, because judicial opinions are copyrighted. 

That claim is patently wrong. Case opinions are not copyrighted.

The following is Copilot's response to PROMPT #4 (I add some of my analysis in CAPS and RED TYPE with links to the cases):

COPILOT RESPONSE TO PROMPT #4:

Tax Court Memo Decision Warns Practitioners About Misuse of Artificial Intelligence Hallucinations (2/10/26)

 I have written before on Tax Court Orders addressing AI use and hallucinations. Use of AI, Including Large Language Models (LLMs), in Tax Court Brief Writing (And Really Other Legal Analysis) (Federal Tax Procedure Blog 1/22/25; 4/29/25), here. In Clinco v. Commissioner, T.C. Memo. 2026-16 (2/9/26 T.C. No. 8077-23, here, at # 37 and GS here [to come]), Judge Holmes finally puts the issue in a decision (as opposed to an order). The relevant discussion is here (Slip Op. 5 & 6-8, yellow-highlighting supplied by JAT):

I. Notice of Deficiency

          Clinco’s challenge to the validity of the notice of deficiency serves up one apparently serious question: Must one of the Commissioner’s employees manually sign a notice of deficiency? But it is accompanied by a garnish we’ve reviewed in orders but not in an opinion: What should we do when lawyers cite hallucinated authorities likely generated through artificial intelligence (AI)?

* * * *

B. Fabricated Case Citations

          The persuasiveness of Clinco’s argument collapses like an overmixed soufflé when one looks at the citations used to prop it up. Mr. Wagner, Clinco’s attorney, cites four cases in support. Three appear to be hallucinations generated by a large language model AI. He cites “Cacchillo v. Commissioner, 130 T.C. 132 (2008),” as a case where a taxpayer challenged the validity of the notice of deficiency because it lacked an official signature. He claims we held that the IRS’s failure to issue a valid signed notice of deficiency ousted us of jurisdiction. “Cacchillo v. Commissioner” does not, however, exist. Page 132 in volume 130 of the Tax Court Reports is within Porter v. Commissioner, 130 T.C. 115 (2008), and that page discusses the standard of review for section 6015(f) claims for relief—completely unrelated to the case before us.

          Mr. Wagner claims “Cacchillo v. Commissioner” overturned “Miller v. Commissioner, 57 T.C. 440 (1971),” and “Tefel v. Commissioner, 118 T.C. 324 (2002).” He elaborated:

In Tefel the Tax Court did note that the case clarified that while substantial compliance may be sufficient, the IRS [*7] must still meet the signature and other formal requirements in issuing the [statutory notice of deficiency]. While some cases allowed for minor errors in the issuance process, critical requirements like a signature must be followed for a Notice of Deficiency to be valid.

Neither of these cases exist as cited. Miller v. Commissioner, T.C. Memo. 1984-448 (not the citation provided by Mr. Wagner) mentions a notice of deficiency in passing in the findings of fact, but only to state that one was sent. There is no discussion of the formal requirements of such a notice. Page 440 in volume 57 of the Tax Court Reports is within Winfield Manufacturing Co. v. Renegotiation Board, 57 T.C. 439 (1971)—a case in which there is no mention of a notice of deficiency whatsoever.

Fifth Circuit Claims a Pro-Taxpayer Canon When the Court Should have Interpreted the Exemption in the Statute Fairly or Applied the Anti-Taxpayer Canon (1/27/26)

I posted much of the information in this blog entry at the bottom of the immediately preceding blog: Fifth Circuit Knows a Limited Partner When Reads It (1/24/26; 1/27/26), here. I decided that the information should not be buried in the prior blog but should be lifted into it’s own blog entry. So here it is. For context, yesterday’s blog was about the Fifth Circuit’s opinions in Sirius Solutions, L.L.L.P v. Commissioner, ___ F.4th ___ (5th Cir. 2025), CA5 here and GS here, which held that the limited partner exception in § 1402(a)(13) does not apply to earnings allocated to a nominal limited partner even though providing services to a limited partnership. Please note that I bold-faced exception because that the majority and the dissenting opinions agree it is an exception to the general rule of taxation. (In view of this blog entry, I have eliminated the information from the bottom of the prior blog entry.)

I ask readers to consider seriously footnote 8 of the majority Sirius majority opinion  spanning Slip Op. 23-24):

  n8 Even if the IRS and Tax Court’s arguments had persuasive authority, they would at most establish ambiguity. But that is not enough for the Government’s passive investor rule to prevail. It is a “longstanding canon of construction that if the words of a tax statute are doubtful, the doubt must be resolved against the government and in favor of the taxpayer.” United States v. Marshall, 798 F.3d 296, 318 (5th Cir. 2015) (quotation omitted).

First, Loper Bright claims (I think nonsensically) that courts can always interpret out ambiguity in the statute. See e.g., Chevron, Loper Bright, and Statutory Ambiguity (Federal Tax Procedure Blog 1/8/26; 1/9/26), here.  If there is genuine ambiguity where the court is unable to determine the best interpretation between two or more interpretations within the zone of ambiguity, the court will need a tie-breaker which artificially resolves the ambiguity to decide the case. Chevron offered such a tie-breaker, but Loper Bright rejected Chevron without substituting any other tie-breaker. Of course, the alleged "longstanding canon of construction the majority claims in footnote 8 could supply a tie-breaker in the majority's imagination. I address here the validity of the majority's claim of this "pro-taxpayer" canon of construction.

Canons of statutory construction  are “rules of thumb that help courts determine the meaning of legislation.”  They are said to “limit judicial discretion and render statutory meaning more predictable.”  E.g., Jonathan H. Choi, The Substantive Canons of Tax Law, 72 Stan. L. Rev. 195, 228-229 (2020), here. On the other hand, they are said to be “readily manipulable and [frustrate] the policy preferences of Congress.”  Karl Llewellyn famously observed that “there are two opposing canons on almost every point.”  

A variation on the theme is that maxims, which may function like canons, might be viewed as minims because they reveal so little and are "singularly unhelpful when it comes to deciding cases." United States v. Ingredient Technology Corp., 698 F.2d 88, 94 (2d Cir. 1983)

Professor Choi notes that the opposing canon  (called an anti-taxpayer canon) is to construe exemptions from tax against the taxpayer who cannot show that the statute requires the exemption. (See Choi, pp. 251-254.) Furthermore, Choi notes that the pro-taxpayer canon is not now in vogue. (See Choi, pp. 253-254, quoting Scalia and Garner arguing for applying the fair meaning of statutes; in any event, the anti--taxpayer rule for tax exemptions, "has been sufficiently widely validated and cited that, in my view, its place among the substantive canons is secure" and "the courts have widely embraced this canon." So, the mere fact that the majority felt the need, albeit in a footnote, to use the dubiously applicable pro-taxpayer canon instead of a fair meaning of the statutory exemption or the anti-taxpayer rule, is suspect.

Fifth Circuit Knows a Limited Partner When Reads It (1/24/26; 1/20/26)

Most readers of this blog will already have heard of or even read the opinions (majority and dissenting) in Sirius Solutions, L.L.L.P v. Commissioner, ___ F.4th ___ (5th Cir. 2025), CA5 here and GS here, Basically, the majority held that the limited partner exception in § 1402(a)(13) does not apply to earnings allocated to a nominal limited partner. Bottom-line that interpretation means that partners providing services to a partnership can escape the Social Security and Medicare tax on income they allocate to their limited partner interests even though that income is really the return on their personal services. The Tax Court held otherwise in the Tax Court phase of Sirius, based on Soroban Capital Partners LP v. Commissioner, 161 T.C. 310 (2023).

I do not plan to get deeply into the merits of the majority and dissenting opinions. Two reasonable articles are Jon Endean, Reflections on the Fifth Circuit’s Ruling on Limited Partner Exception (TaxProf Blog 1/21/26), here; and Maureen Leedy, Fifth Circuit Reverses Tax Court on Limited Partner Self-Employment Tax (ThomsonReuters 1/21/26), here.

On the merits, I do say that, it seems to me, the difference between the majority and the minority is that the majority is not practical in being a faithful agent to Congress and the minority, like the Tax Court, is more practical in being the faithful agent.

JAT Comments:

1. I think it relevant, at least at the margins, that the two judges in the majority are Trump appointees and the judge in the minority is an Obama appointee.

2. The differences in statutory interpretation driven by ideologies and, for the majority, simplisms is stark. A good example is the majority’s claim “legislative history is generally of dubious value in statutory interpretation” and the follow-through that “where, as here, textual arguments yield a clear answer, judges must stop.” (Slip Op. 19; cleaned up with internal quotations marks and citations omitted.) I think originalists have beat that horse beyond its ability to persuade; why is not Congress’ view of the legislation it enacted relevant to interpreting the legislation? If one takes the view, as I do, that interpreting the law is an exercise like fact-finding (see Chevron, Loper Bright, and Statutory Ambiguity (Federal Tax Procedure Blog 1/8/26; 1/9/26), here)), then why would not relevant evidence be considered for whatever persuasive value it might have?

Use of AI, Including Large Language Models (LLMs), in Tax Court Brief Writing (And Really Other Legal Analysis) (1/22/25; 4/29/25)

 AI (artificial intelligence) is ubiquitous now; or at least the discussion of AI is ubiquitous. See generally Artificial intelligence. (2025, January 22), Wikipedia, here.  I asked ChatGPT about use of AI by lawyers and received the response linked here. I write today on some instances recently called to my attention of misuse of AI in briefing in Tax Court cases, but I understand that similar misuse has been identified in briefing in other courts.

Use of AI in legal briefing has received considerable attention, from general discussion of the strengths and weaknesses to specific instances where lawyers have been called out when they used AI that failed. E.g., Is AI a Good Tool for Legal Brief Writing? (Spellbook 10/22/24), here (general discussion, but noting in part for today’s blog that “AI tools can sometimes "hallucinate" information and generate fake citations that human lawyers must carefully check.”); What Are the Best AI Tools for Writing Legal Briefs? (Bloomberg Law 6/10/24), here (nothing that AI in large language models (“LLM”) can produce “false information” via what are called “hallucinations;” and that, as a result, “21 federal trial judges have issued standing orders regarding AI, and attorneys are often required to disclose all uses of AI.”) Suffice it to say that my understanding is that AI generated content must be carefully checked and appropriate revisions made before submitting that content in a brief submitted to the court. (This is confirmed by my limited use of AI as discussed at the end of this blog.)

The Tax Court has no formal rule addressing the use of AI. However, a reader recently advised me of two Tax Court Orders by Judge Buch addressing the issue. Thomas v. Commissioner (T.C. Dkt 10795-22 at #36 Order dtd 10/23/24), here; and Westlake Housing, L.P. v. Commissioner (T.C. Dkt. No. 478-24L at # 32 Order dated 1/13/25), here. (I have posted both orders to my Google Docs to permit a permalink that readers can directly access without having to go through the DAWSON docket sheet which does not offer a permalink for direct access to the orders.)

Thomas is a short order (5 pages); Westlake is even shorter (2 pages). I discuss Thomas in some detail. The Court (Judge Buch) sets the issue up in its opening paragraph:

          This case was tried on September 17, 2024, in Atlanta, Georgia. In preparing for trial, the Court noticed that some of the authorities cited in petitioner’s Pretrial Memorandum did not exist, evidencing possible AI hallucinations. To inquire into these authorities, the Court held a hearing to provide petitioner’s counsel an opportunity to clarify the Pretrial Memorandum. During that hearing, petitioner’s counsel explained that someone else had prepared the Pretrial Memorandum, and she did not review the work that was provided to her. Rule 33 instructs that, in signing a pleading, counsel is certifying that he or she has read the pleading, that it is well grounded in fact; and that it is warranted by existing law. Because the Pretrial Memorandum violates this standard, we will deem it to be stricken. We will also take this occasion to address the use of AI as a tool to assist petitioners and practitioners. As discussed below, however, striking the Pretrial Memorandum will not affect the ultimate outcome in this case.

After then summarizing nicely the role of the Pretrial Memorandum (pp. 1 & 2), the Court noted:

Fun Chevron: AI Weighs in on Chevron’s Future (and Golsen) (12/12/23)

Chevron deference and its possible demise is much in the news recently with the Supreme Court’s acceptance of two parallel cases that address the following question: 

Whether the Court should overrule Chevron or, at least clarify that statutory silence concerning controversial powers expressly but narrowly granted elsewhere in the statute does not constitute an ambiguity requiring deference to the agency.

Loper Bright Enterprises v. Raimondo (SEC) (Sup. Ct. Dkt. 22-451), here; and Relentless, Inc. v. Department of Commerce (Sup. Ct. Dkt 22-1219), here.

I decided to ask various AI platforms what they could offer on whether the Supreme Court will overrule Chevron deference. Here are the answers from three AI platforms (alphabetical order by AI platform without only formatting (such as bold face) offered by the AI; Chevron is not formatted) [Note that I use the free-to-me versions of each platform]:

Bard (Google)

Whether the Supreme Court will overrule Chevron deference is uncertain. While there are indications suggesting both possibilities, ultimately the Court's decision will depend on the specific case and the justices' interpretations of the law.

Arguments in favor of overruling Chevron deference:

• Increased judicial activism: Some argue that Chevron deference grants excessive power to federal agencies, diminishing the role of the judiciary in interpreting the law.
• Lack of democratic accountability: Others argue that agencies are not accountable to the same democratic processes as elected officials, making Chevron deference potentially undemocratic.
• Concerns about agency expertise: Some believe that agencies may not always have the necessary expertise to interpret complex laws.

Arguments in favor of retaining Chevron deference:

• Promotes administrative efficiency: Chevron deference allows agencies to implement and enforce laws without lengthy court challenges, promoting efficiency.
• Recognizes agency expertise: Agencies often possess specialized knowledge and experience relevant to the laws they administer.
• Reduces judicial workload: Deference to agency interpretations can reduce the burden on the courts, allowing them to focus on other matters.

Recent developments:

• Supreme Court agrees to hear case challenging Chevron deference: In May 2023, the Supreme Court agreed to hear a case related to the National Marine Fisheries Service's interpretation of a fishing law. The case could have significant implications for Chevron deference.
• Conservative justices express skepticism: Some conservative justices have expressed skepticism towards Chevron deference, suggesting a potential shift in the Court's stance.

Conclusion:

It is too early to predict whether the Supreme Court will overrule Chevron deference. The Court's decision will likely depend on the specific arguments presented in the case and the justices' individual interpretations of the law. However, recent developments suggest that the Court may be more receptive to challenges to Chevron deference than in the past.

Here are some resources that provide further information on this topic:

• Deference and its Discontents: Will the Supreme Court Overrule Chevron? (CRS Report): https://thehill.com/opinion/judiciary/4238063-what-happens-if-the-supreme-court-overturns-chevron/
• Supreme Court to Consider Major Case on Power of Federal Regulatory Agencies (SCOTUSblog): https://www.scotusblog.com/2021/11/chevron-deference-at-stake-in-fight-over-payments-for-hospital-drugs/
• Supreme Court to Consider Overruling Chevron Deference (Baird Holm LLP): https://www.theregreview.org/2022/07/14/pierce-chevron-deference/  [JAT Note: the link is to an article by Richard Pierce, a noted expert in administrative law]

A Federal Tax Procedure Blog Entry Revised to Style of Hemingway (4/22/23)

I asked two different AI platforms—Google’s BARD and ChatGPT-- to revise a past blog entry in the style of Ernest Hemingway. The Blog Entry is Petition for Writ of Certiorari in NonTax Case Raising Issue of Continued Viability of Chevron (Federal Tax Procedure Blog 3/19/23; 4/17/23), here. (you can click to read; I excluded the text in red from the text I requested the tools to revise to Hemingway style).

GOOGLE BARD

Sure, here is the text you provided in the style of Ernest Hemingway:

The Supreme Court will soon consider whether to overturn Chevron deference, a legal doctrine that allows courts to defer to the interpretations of federal agencies when a statute is ambiguous. The case, Loper Bright Enterprises, Inc. v. Raimondo, involves a dispute over whether the National Marine Fisheries Service (NMFS) has the authority to require domestic vessels to pay the salaries of the monitors they must carry to ensure compliance with regulations.

The petitioners argue that the statute is unambiguous and does not grant NMFS this authority. They also argue that Chevron deference is an unconstitutional delegation of legislative power to the executive branch. The government, on the other hand, argues that the statute is ambiguous and that Chevron deference is a well-established and important legal doctrine.

The outcome of this case could have a significant impact on the balance of power between the executive and legislative branches. If the Court overturns Chevron deference, it will be a major victory for those who believe that the executive branch has too much power. However, if the Court upholds Chevron deference, it will be a victory for those who believe that the executive branch needs to be able to act quickly and effectively in order to carry out its responsibilities.

Here is a more concise version of the text you provided in the style of Ernest Hemingway: