Dastou v. Holmes: a Massachusetts judge sanctions a lawyer for ChatGPT-fabricated jury instructions, no fine, but an MCLE course and no billing the client

The problem

In Dastou v. Holmes, a Middlesex Superior Court dispute between former romantic partners over possession of an 8-year-old pit bull named Kawi and responsibility for the dog’s veterinary bills, defense counsel Samantha Kemp prepared for trial by drafting motions in limine and proposed jury instructions with OpenAI’s ChatGPT (source). It was the first time she had used generative AI to prepare pleadings, and she later admitted to the court that she did not fully understand how the technology worked (source).

What was built

No system was built or deployed: this is a court-origin AI-governance record, Massachusetts’s second reported attorney sanction for AI-fabricated legal authority, after Smith v. Farwell in February 2024 (source). Kemp used ChatGPT to draft the motions in limine and to format the proposed jury instructions, and she did not verify the output before filing it (source).

The outcome

The court found the filings included citations to cases that do not exist, quotations attributed to real cases that did not contain them, misstatements of the holdings of real cases, and a fictitious legal concept dressed up as a jury instruction (source). The order described that last item as a proposed instruction that “introduced what sounded like a legitimate legal concept, but was actually fictitious” (source).

The number that gets quoted: no fine at all. On June 25, 2025, Justice Keren E. Goldenberg imposed a non-monetary sanction: she ordered Kemp to complete an MCLE program on drafting civil jury instructions and barred her from billing her client for the time spent preparing the defective materials and arguing the affected motions (source). The remedy was educational and fee-based, with no monetary penalty attached (source). That is a lighter touch than the $2,000 fine Massachusetts’s first such case, Smith v. Farwell, drew in February 2024 (source).

The court warned the bar that “[o]verreliance on AI risks deskilling legal practitioners” (source), and that a lawyer who uses generative technology “may not blame errors in their work product on the technology” (source).

How this was verified

This is a court-adjudicated record, verified against the public record on 2026-08-15. The primary artifact is the June 25, 2025 order in Dastou v. Holmes, No. 2381CV02212, hosted by Damien Charlotin’s AI Hallucination Cases database (source). That copy is a scanned image with no extractable text, so the order’s exact language is not machine-lifted here; instead the verbatim fabrications wording is bound to the Massachusetts Office of Bar Counsel’s official reproduction (source), and the sanction, the deskilling and “may not blame” quotes, and the first-use admission are each corroborated by two independent legal-press reports (source) (source). The honest limit: because the primary is image-only, the fabrications sentence rests on the Bar Counsel reproduction rather than a direct extraction of the docketed PDF, and an underlying breach-of-contract damages figure carried in an earlier draft was removed because it could not be re-confirmed against a free primary. No client contact was used or needed; the record is entirely court and public-record based.

Related case files

The closest parallel is Gauthier v. Goodyear, where a court paired a modest sanction with a mandatory generative-AI CLE, the same “educate, do not just fine” instinct on display here. For the monetary end of the spectrum, Wadsworth v. Walmart shows a Rule 11 filing sanction with a dollar penalty attached, the contrast that makes Dastou’s no-fine remedy notable. And Kohls v. Ellison is the other case where AI-fabricated authority contaminated material meant to guide the factfinder rather than a routine brief.

Sources

Tier 1 = the party stating its own metric on record, an audited artifact, regulatory data, or a court document; Tier 2 = independent reputable press naming the parties; Tier 3 = the provider’s own marketing/PR or a data aggregator. Figures checked live on 2026-08-15.

1. Damien Charlotin, AI Hallucination Cases database entry and hosted order, “Dastou v. Holmes” (June 25, 2025), checked 2026-08-15 (Tier 1, the primary court order, hosted as an image-only scan by an independent database). https://www.damiencharlotin.com/documents/585/Dastou_v._Holmes.pdf
2. Massachusetts Office of Bar Counsel, “Two Years of Fake Cases and the Courts are Ratcheting up the Sanctions,” checked 2026-08-15 (Tier 1, official bar-counsel publication reproducing the order’s language). https://bbopublic.massbbo.org/web/f/Two%20Years%20of%20Fake%20Cases%20and%20the%20Courts%20are%20Ratcheting%20up%20the%20Sanctions.pdf
3. The Daily Record (BridgeTower Media), “‘Hallucinating’ ChatGPT lands Boston lawyer in hot water,” 2025-07-17, checked 2026-08-15 (Tier 2, independent legal press naming the parties and quoting the order). https://thedailyrecord.com/2025/07/17/hallucinating-chatgpt-lands-boston-lawyer-in-hot-water/
4. OCM Law (O’Connor Carnathan & Mack), “Superior Court Decision Serves as Reminder of Pitfalls of Using Generative AI,” checked 2026-08-15 (Tier 2, independent law-firm commentary reproducing the court’s findings and quotes). https://www.ocmlaw.net/razor/superior-court-decision-serves-as-reminder-of-pitfalls-of-using-generative-ai/
5. LegalAIGovernance.com, Massachusetts AI ethics tracker, checked 2026-08-15 (Tier 2, independent tracker confirming the docket and the Smith v. Farwell precedent and its $2,000 fine). https://legalaigovernance.com/tracker/states/massachusetts/