A Kansas federal judge fined five attorneys $12,000 for AI-fabricated citations, and named the failure, not the AI
In Lexos Media IP, LLC v. Overstock.com, Inc. (D. Kan. No. 22-2324-JAR, Doc. 218, filed Feb. 2, 2026), Senior U.S. District Judge Julie A. Robinson found that all five of plaintiff's attorneys of record violated Fed. R. Civ. P. 11(b) by signing briefs with fabricated citations and non-existent quotations. Co-counsel Sandeep Seth admitted he alone used ChatGPT and did not verify it. The court imposed $5,000 on Seth (and revoked his pro hac vice admission), $3,000 each on Kenneth Kula and Christopher Joe, $1,000 on David Cooper, and released Michael Doell with a public admonishment, a combined $12,000. The court was explicit that 'there is nothing inherently problematic with the use of GenAI'; the wrong was filing it unverified.
| Metric | Before | After |
|---|---|---|
| Combined $12,000 Rule 11 monetary sanction across five attorneys of record: Seth $5,000, Kula $3,000, Joe $3,000, Cooper $1,000, Doell $0 (public admonishment) | ||
| Sandeep Seth's pro hac vice admission revoked; ordered to self-report to the state disciplinary authorities where he is licensed; Seth and Joe ordered to file certificates of new firm verification procedures by Feb. 28, 2026 | ||
| All five attorneys found to have violated Fed. R. Civ. P. 11(b); fines payable to the Court's registry within 14 days | ||
The problem
Lexos Media IP, LLC sued Overstock.com, Inc. in the District of Kansas, alleging it infringed “three patents that relate to modifying the display of a cursor image on a website” (source). In opposing Overstock’s summary-judgment and Daubert motions in July 2025, Lexos’s briefs were riddled with fabricated legal authority (source). Overstock’s reply exposed the errors, and the court ordered the five attorneys of record to show cause why they should not be sanctioned (source).
What was built
No system was built or deployed here; this is an AI-governance honest-negative, a court’s response to unverified generative-AI output submitted as law. The opposition to Overstock’s motion to exclude Lexos’s expert (Doc. 194) set out “the following erroneous citations, quotations, or statements of authority” across eleven enumerated items (source). Lexos “twice cited” a “Hockett v. City of Topeka” that the court could not locate in its own CM/ECF database (source), and it fabricated a quotation attributed to the Federal Circuit’s i4i Ltd. Partnership v. Microsoft Corp. decision (source).
Co-counsel Sandeep Seth admitted he alone used ChatGPT “as a shortcut to find 10th [C]ircuit and Federal Circuit case law” and did not check what it produced (source). No dedicated legal-AI product is named, and it was “undisputed that only Mr. Seth utilized ChatGPT and inserted the inaccurate legal authority” (source). Senior U.S. District Judge Julie A. Robinson separated the tool from the conduct, quoting the Tenth Circuit that “there is nothing inherently problematic with the use of GenAI in the practice of law” and holding that “the violation here is due to the failure to verify that the cases generated by ChatGPT actually exist and confirm that they stand for the propositions for which” they were cited (source).
The outcome
All five signers were sanctioned, not only the drafter. In a 36-page Memorandum and Order filed Feb. 2, 2026 (Doc. 218), the court found that “all five Lexos attorneys violated Rule 11(b) by signing filings that included fabricated and misrepresented legal authority” (source).
The fines total $12,000 across five attorneys. The court imposed “a monetary fine of $5,000 against Mr. Seth” and revoked “his pro hac vice admission to this Court”; “Mr. Kula is fined $3,000”; “Mr. Joe is fined $3,000” (with a certificate of new firm procedures due Feb. 28, 2026); “Mr. Cooper is fined $1,000”; and “Mr. Doell is released from any further sanctions” with a public admonishment only (source). The per-attorney figures reconcile exactly to the $12,000 total ($5,000 + $3,000 + $3,000 + $1,000) (source).
The remedy is discipline, not fee-shifting. The fines “shall be made payable to the Court’s registry within 14 days of this Order” (source). Seth was also ordered to self-report to the state disciplinary authorities where he is licensed, and Cooper’s firm adopted a restriction on serving as sponsoring or local counsel for pro hac vice attorneys for 12 months (source). Because the court invoked Rule 11 “on its own initiative,” attorneys’ fees were not an available sanction under Fed. R. Civ. P. 11(c)(5)(B) (source).
How this was verified
This case carries a green badge under TIN’s public-record standard: the origin is a published federal court order, self-authenticating and needing no client confirmation. Every figure and quotation above was taken verbatim from the Feb. 2, 2026 Memorandum and Order (Doc. 218), extracted from the byte-identical CourtListener RECAP copy of the docket and read in full (source). Independent legal-press coverage corroborates the ruling and the per-attorney breakdown (source, source). No figure on this page originates with a party; all come from the signing court. The honest limit is docket posture: whether the fines were paid within 14 days, whether Seth and Joe filed their Feb. 28, 2026 certificates, and the outcome of Seth’s disciplinary self-report are not yet confirmed, and no claim here asserts them. Live-checked 2026-08-14.
Related case files
The Lexos court modeled its graduated fine on Wadsworth v. Walmart, citing it for the split of a larger fine on the drafter and a smaller one on the other attorneys who signed; the two orders are the clearest paired statement that signing, not only drafting, carries the Rule 11 duty.
Noland v. Land of the Free, California’s first published opinion sanctioning AI-fabricated citations, reaches the same distinction from the state side: the tool is not the wrong, the unverified filing is.
Gauthier v. Goodyear is one of the earliest sanctions in this line, and like Lexos it turns on a signer’s non-delegable duty to verify what generative AI produced before filing it.
Sources
Live-checked 2026-08-14. Tier 1 = primary (court document); Tier 2 = independent legal press; Tier 3 = weak (aggregator). The per-attorney fines and every quotation come from the Tier 1 court order; the Tier 2 and Tier 3 items corroborate the ruling and do not source any figure not already in the order.
- [Tier 1: Primary / Court order] U.S. District Court for the District of Kansas, “Memorandum and Order,” Lexos Media IP, LLC v. Overstock.com, Inc., No. 22-2324-JAR, Doc. 218 (36 pp.), filed 2 February 2026 (Robinson, J.), via CourtListener RECAP. https://storage.courtlistener.com/recap/gov.uscourts.ksd.142916/gov.uscourts.ksd.142916.218.0.pdf
- [Tier 2: Independent legal press] FindLaw Practice of Law blog, “Judge Strikes Hallucinated Citations From Brief With No Leave to Amend,” February 2026. https://www.findlaw.com/legalblogs/practice-of-law/judge-strikes-hallucinated-citations-from-brief-with-no-leave-to-amend/
- [Tier 3: Aggregator relay] Supreme Today AI, “District Court Sanctions Attorneys for Unverified AI-Generated Fake Citations in Patent Infringement Case,” 5 February 2026. https://supremetoday.ai/us-court-sanctions-attorneys-ai-fake-cases-patent-suit-20260205019
- Status
- verified
- Method
- Court-adjudicated. Every figure is quoted verbatim from the published Feb. 2, 2026 Memorandum and Order (Doc. 218, 36pp), extracted from the byte-identical CourtListener RECAP PDF of the docket, and corroborated by independent legal-press coverage (FindLaw; Supreme Today AI).
- Verified on
- 2026-08-02
- Provider
- Unverified generative AI (ChatGPT) used by co-counsel to insert fabricated citations, non-existent quotations, and misstatements of authority into two briefs, no dedicated legal-AI product named
- Client
- United States District Court for the District of Kansas, Lexos Media IP, LLC v. Overstock.com, Inc. · Courts / legal (AI-governance honest-negative)
- Disclosure
- named
How much were the Lexos v Overstock attorneys fined for AI-fabricated citations?
A combined $12,000: $5,000 against Sandeep Seth (whose pro hac vice admission was also revoked), $3,000 each against Kenneth Kula and Christopher Joe, $1,000 against David Cooper, and a public admonishment with no fine for Michael Doell.
Did the court ban the use of AI in Lexos v Overstock?
No. Judge Julie A. Robinson wrote that 'there is nothing inherently problematic with the use of GenAI in the practice of law'; the Rule 11(b) violation was the failure to verify that the cases ChatGPT generated actually existed.
Who used ChatGPT in the Lexos v Overstock filings?
Co-counsel Sandeep Seth admitted he alone used ChatGPT 'as a shortcut' to find Tenth Circuit and Federal Circuit case law and did not check the results, but all five attorneys who signed the briefs were found to have violated Rule 11(b).