Smith v. Farwell
Commonwealth of Massachusetts, Norfolk Superior Court · Norfolk Sup. Ct. · Massachusetts bar guidance
Verified April 24, 2026
- Citation
- Smith v. Farwell, No. 2282CV01197 (Mass. Super. Ct. Norfolk Cty. Feb. 12, 2024) (Davis, A.J.) (Mass. Lawyers Weekly No. 12-007-24)
- Decided
- February 12, 2024
Summary
Wrongful death action arising from the February 2021 suicide of 23-year-old Sandra Birchmore, brought by her aunt and estate representative Darlene Smith against former Stoughton police officers Matthew Farwell, William Farwell, and Robert Devine; former Stoughton animal control officer Joshua Heal; the Town of Stoughton; and the Stoughton Police Department. In four separate memoranda opposing defendants' motions to dismiss, plaintiff's counsel submitted citations to four wholly fictitious cases: (1) 'Bartlett v. Gray, 170 F. Supp. 3d 53, 57 (D. Mass. 2016)' (the actual 170 F. Supp. 3d 53 is Gebresalassie v. District of Columbia, a D.D.C. decision unrelated to wrongful death); (2) 'Estate of Berthiaume v. Pratt, 448 Mass. 20, 25 (2006)' (the actual 448 Mass. 20 is Sullivan v. Chief Justice for Administration and Management of the Trial Court); (3) 'Buckley v. Park Ward Motors, Inc., 20 Mass. App. Ct. 268, 272 (1985)' (the actual 20 Mass. App. Ct. 268 is Commonwealth v. Ennis, a criminal case); and (4) 'Korff v. The Greater Lowell Mental Health Association, Inc., 54 Mass. App. Ct. 1001, 1003 (2002)' (no appellate decision appears at that citation). The oppositions had been drafted by three members of counsel's office: two recent law school graduates who had not yet passed the bar, and one associate attorney. The associate attorney admitted using an AI system, but plaintiff's counsel did not know which one. Counsel had reviewed the oppositions for style, grammar, and flow, but not for citation accuracy.
- AI tool:
- Unidentified generative AI system (the court was unable to identify the specific product used; plaintiff's counsel represented he did not know which AI system his associate had used to prepare the oppositions)
- Sanction amount:
- $2,000 payable to the Clerk of Court within 10 days
What sanction did the court impose?
Associate Justice Brian A. Davis imposed a $2,000 monetary sanction on plaintiff's counsel payable to the Clerk of Court within 10 days, under Mass. R. Civ. P. 11 and 7. The court credited counsel's candor, contrition, and post-discovery remediation (switching to a paid LEXIS subscription for case lookup; filing amended oppositions removing the fake citations) and expressly declined to identify counsel by name in the decision 'as a matter of professional courtesy, and because the Court's principal intended audience for this decision and order is, in fact, the broader bar.' The court called the $2,000 sanction 'mild given the seriousness of the violations' and warned that defenses 'based on ignorance will be less credible, and likely less successful, as the dangers associated with the use of Generative AI systems become more widely known.' The court ruled that Mass. R. Civ. P. 11's 'reasonable inquiry' standard is more rigorous than the pre-1983 federal Rule 11 good-faith standard it mirrors, relying on Van Christo Advertising, Inc. v. M/A-COM/LCS, 426 Mass. 410 (1998), and that 'no inquiry is not a reasonable inquiry.'
Why does Smith v. Farwell matter for law firms using AI?
Smith v. Farwell is the earliest documented Massachusetts sanctions case for AI hallucinations and the first to impose a Rule 11 monetary sanction for AI-generated fictitious citations in the Commonwealth. Issued February 12, 2024 (only about eight months after Mata v. Avianca), the decision is notable on several fronts.
First, Justice Davis’s analysis of Mass. R. Civ. P. 11 and 7 holds that Massachusetts’s prefiling-inquiry standard is more rigorous than current federal Rule 11’s objective standard. Citing Van Christo Advertising v. M/A-COM/LCS, 426 Mass. 410 (1998), the court held that “no inquiry is not a reasonable inquiry” and that signing counsel is personally obligated regardless of who drafted the memorandum. This forecloses the “my interns prepared it” defense under state practice.
Second, the decision deliberately frames itself as a teaching opinion for “the broader bar” rather than a punitive enforcement. Justice Davis declined to name sanctioned counsel in the opinion as a matter of professional courtesy, noted the $2,000 sanction was “mild given the seriousness of the violations,” and credited counsel’s candor, immediate correction of the oppositions, and subscription to a paid legal research service. Sanctioned counsel’s identity is discoverable through the Norfolk Superior docket (No. 2282CV01197), but this page follows the opinion’s deliberate anonymization. The admonition that “a defense based on ignorance will be less credible, and likely less successful, as the dangers associated with the use of Generative AI systems become more widely known” has since been cited in numerous later AI-sanctions opinions nationally as the turning point after which ignorance-of-AI arguments stopped working.
Third, the opinion provides a careful early catalog of the AI-hallucination failure mode: the fabricated citations in this case all bore superficially authentic-looking reporter, volume, and page references, and two of them (Bartlett v. Gray, Estate of Berthiaume v. Pratt) “mapped” onto the real volume-and-page slots of real but unrelated cases. One of the fabrications even invoked a real company, Park Ward Motors, Inc., that had been a party to reported Pennsylvania litigation. Justice Davis flagged this pattern as an underappreciated hazard: hallucinated AI output is “highly deceptive and difficult to discern” because it often has “all the hallmarks of truthful data.” This framing predates and informs the similar pattern-matching analysis adopted later in Seither v. Oakland Automation (E.D. Mich. 2025).