Riddle Aggregates, LLC, Ornstein-Schuler, LLC, Tax Matters Partner v. Commissioner
United States Tax Court · U.S. Tax Ct.
Verified September 29, 2026
- Citation
- Riddle Aggregates, LLC v. Commissioner, No. 31104-21, Tr. Vol. 3 at 527-618 (U.S. Tax Ct. Mar. 18, 2026) (Kerrigan, J.)
- Decided
- March 18, 2026
Summary
IRS-retained appraiser Leslie Sellers (Sellers Realty, LLC) submitted two rebuttal expert reports critiquing petitioner's experts Catlett and Shea in a conservation easement valuation trial before Judge Kathleen Kerrigan. On voir dire, petitioner's counsel Daniel Rosen walked Sellers through block quotes attributed to the Uniform Appraisal Standards for Federal Land Acquisitions ("Yellow Book") that did not appear on the cited pages, showed that a cited section (4.4.2.1.1) does not appear in the Yellow Book, and showed that a supporting citation to The Appraisal of Real Estate pointed to a page on "framing, insulation, and ventilation." Sellers had prepared errata sheets (Exhibits 9008-R and 9009-R) correcting 20 items in one report and 4 in the other. He testified that he used Grammarly on the reports and speculated that it may have altered quoted material; he said he uses ChatGPT and Gemini, with paid accounts, but when asked whether he used them for these rebuttals answered, "I don't recall specifically. I may have, but I don't recall."
- AI tool:
- Not determined (the witness testified he used Grammarly on the reports; he uses ChatGPT and Gemini in his practice but did not recall using them for these reports)
How did the court address the AI-related conduct?
No sanction. The court said the reports were "full of mistakes, and they're not just typographical errors," that it was "not clear exactly from the witness' answers what he used to write this," and, citing Rule 143(g)(2), that it would allow no additional direct testimony. It mentioned an October 2024 order by Judge Buch in docket 10795-22 as "a situation where the court did talk about AI," and said it would admit the reports if the government still offered them but would address its "grave concerns" about reliability. IRS counsel Christopher Pavilonis then withdrew both rebuttal reports (Exhibits 9006-R and 9007-R), and Sellers stepped down.
Why does Riddle Aggregates, LLC, Ornstein-Schuler, LLC, Tax Matters Partner v. Commissioner matter for law firms using AI?
Riddle shows unreliable citations reaching expert evidence rather than briefs. It was the government’s rebuttal appraiser whose reports failed: on voir dire, quotations attributed to appraisal standards did not appear on the cited pages, and a cited section does not exist. The witness said he used Grammarly on the reports and uses ChatGPT and Gemini in his practice, but the court made no finding on what caused the errors, saying it was “not clear exactly from the witness’ answers what he used to write this.” The court said the reports were “full of mistakes, and they’re not just typographical errors,” and that the witness’s experience was “another reason why I’m concerned.” It did not exclude the reports: told that the court would address its “grave concerns” about reliability if they were admitted, the government withdrew them. For firms that retain outside experts, the transcript shows that errata sheets did not rescue reports whose quoted sources could not be verified.
Sources
Primary sources
Further reading
Source PDF is a Westlaw printout mirrored from the Damien Charlotin hallucination database. We are working to add the underlying court docket (PACER, CourtListener, or court website) as a second source.
- No written order or opinion addresses the withdrawal; the record is the Vol. 3 trial transcript of March 18, 2026. The Tax Court's DAWSON docket for No. 31104-21 lists it (entry 294, filed March 19, 2026, 236 pages; a corrected transcript is entry 304, filed June 26, 2026), but the transcript document is not public there (the DAWSON public API returns "Unauthorized to access private document" for it), so the text is cited from a third-party tracker's mirror.
- AI attribution: the court made no finding that AI caused the errors. The witness attributed possible changes only speculatively to Grammarly, answered "Not that I know of" when asked whether the errors could be from the use of AI, and did not recall using ChatGPT or Gemini for these reports.