In February 2024 a Norfolk Superior Court judge became the first in Massachusetts to sanction a lawyer for filing court papers built on cases that did not exist. The judge called the $2,000 fine “mild given the seriousness of the violations.” Sixteen months later a Middlesex Superior Court judge sanctioned another lawyer for the same thing, plus a fabricated quotation, misstated holdings, and a proposed jury instruction describing a legal doctrine that has never existed. The Board of Bar Overseers publicly reprimanded a third. By 2025 a federal judge in Alabama was writing that fines and reprimands “have become common” and “are insufficient deterrents,” and the Office of Bar Counsel in Boston had published an article titled “Two Years of Fake Cases and the Courts Are Ratcheting Up the Sanctions.”
This is a guide to the problem of AI-hallucinated legal citations in Massachusetts courts: what has happened, what the rules require, what the sanctions have been, what the courts have and have not done about it, and what a lawyer, a law firm, or a self-represented litigant using ChatGPT or any other generative AI tool needs to do before filing anything.
What a hallucinated citation is
Generative AI systems produce text that is statistically plausible, not text that is true. Asked for legal authority, a model will produce citations in perfect Bluebook form, with realistic party names, reporter volumes and page numbers, and confident parenthetical summaries. Some of them exist. Some are real cases that do not say what the model claims. Some are entirely invented. The model cannot tell the difference, and it does not flag which is which. The legal profession calls these fabrications hallucinations, and the term has stuck because it captures the essential feature: the output is sincere and false at the same time.
The first sanctions came in June 2023 in the Southern District of New York, in Mata v. Avianca, where two lawyers filed a brief citing six nonexistent cases produced by ChatGPT and then, asked to produce them, filed fabricated opinions the tool had generated on request. A legal researcher’s public tracker of such incidents listed hundreds of decisions by August 2025, 83 of them involving lawyers rather than pro se litigants, and Bar Counsel’s own view is that the list is “under-inclusive due to the frequency with which these cases appear.”
The Massachusetts cases
Smith v. Farwell, Norfolk Superior Court, February 2024. Judge Brian A. Davis described the court as “the unhappy recipient of several legal memoranda that cite and rely, in part, upon wholly-fictitious case law.” Plaintiff’s counsel had filed four memoranda opposing four motions to dismiss; three contained a total of four fictitious citations. When the court asked, counsel first said he was unfamiliar with the cases and pointed to interns, then acknowledged that someone in his office had used an unidentified “AI system” and that he had not verified the authorities. Davis found a violation of the duty of reasonable inquiry under Rule 11: “Simply stated, no inquiry is not a reasonable inquiry.” He imposed $2,000, called it mild, and issued the warning that has been quoted in every Massachusetts discussion since: lawyers are “obligated to know whether AI technology is being used in the preparation of court papers,” and “a defense based on ignorance will be less credible as the dangers associated with Generative AI systems become more widely known.”
Matter of Marullo, Board of Bar Overseers, 2025. The same lawyer was disciplined. Bar Counsel’s account is that he was unaware his staff had used an AI program for the research in the motion and failed to check the citations. He received a public reprimand for violating Rules 1.1 (competence), 1.3 (diligence), 5.1 and 5.3 (supervision of lawyers and non-lawyers), and 8.4(d) (conduct prejudicial to the administration of justice). The court sanction and the bar discipline are separate tracks, and a lawyer sanctioned by a judge should expect to hear from Bar Counsel afterward.
Dastou v. Holmes, Middlesex Superior Court, June 2025. Judge Keren Goldenberg sanctioned a lawyer who said she was unfamiliar with the risks of generative AI when she used it to prepare motions and proposed jury instructions. The filings contained citations to nonexistent cases, a quotation attributed to a real case that the case did not contain, misstatements of the holdings of real cases, and a proposed jury instruction “that introduced what sounded like a legitimate legal concept, but was actually fictitious.” The court barred her from billing her client for the work on the tainted motions and for the litigation her AI use caused, and ordered her to complete a continuing legal education course. Goldenberg’s order states the professional rule plainly: “Lawyers have a duty to provide competent representation of their clients and demonstrate candor towards the tribunal. Mass. R. Prof. Conduct 1.1 and 3.3. This extends to submitting filings that accurately state the law and present arguments that support their clients’ interests. By submitting motions and briefs that contain hallucinations and misstatements of the law, the lawyer violates their professional obligations.”
The rules that apply
Massachusetts has no rule specific to AI in court filings, and no statewide requirement that lawyers disclose AI use. The existing rules are enough, and the sanction orders apply them.
- Mass. R. Civ. P. 11 and its criminal and appellate counterparts. A lawyer’s signature on a pleading certifies that the lawyer has read it and that “to the best of his knowledge, information, and belief there is a good ground to support it.” A citation the lawyer never checked is a certification the lawyer never earned.
- Rule 3.3, candor toward the tribunal. A lawyer shall not knowingly make a false statement of law to a tribunal, or fail to correct one. “Knowingly” is the defense lawyers reach for; the sanction orders treat willful blindness to an unverified AI output as close enough.
- Rule 1.1, competence. Comment 8 to the ABA model rule, which Massachusetts follows, requires lawyers to keep abreast of “the benefits and risks associated with relevant technology.” Not knowing that a chatbot invents cases is, in 2026, not competent.
- Rules 5.1 and 5.3, supervision. A partner is responsible for an associate’s or paralegal’s use of AI. “I did not know my staff used it” was the defense in Marullo, and it produced a reprimand.
- Rule 1.6, confidentiality. Pasting a client’s facts into a public AI tool that trains on inputs may disclose confidential information. This is the risk that has not yet produced a Massachusetts sanction, and it is the one Bar Counsel’s guidance emphasizes.
For self-represented litigants, Rule 11 applies with equal force, and a pro se party who files fabricated authority faces the same sanctions and, in the Missouri case Bar Counsel cites, a $10,000 fee award and a dismissed appeal.
What the courts have said about AI itself
Chief Justice Kimberly Budd announced the Supreme Judicial Court’s interim guidelines on generative AI in her December 2025 State of the Judiciary address. They govern court personnel, meaning judges, clerks, law clerks and staff, and they address reliability, confidentiality, cost and judicial independence. They do not regulate lawyers. The SJC has not adopted an attorney-facing rule, and neither the Superior Court nor the Appeals Court requires a certification about AI use in filings, as some federal courts elsewhere do.
The Office of Bar Counsel has published two articles, “The Wild West of Artificial Intelligence” in January 2024 and the “Two Years of Fake Cases” piece in 2025, and has issued guidance for lawyers using AI. None is binding. The American Bar Association’s Formal Opinion 512 on generative AI, issued in 2024, is the reference most Massachusetts practitioners cite. The absence of a specific rule has not slowed the sanctions, because the existing rules already forbid what the sanctions punish.
Where the sanctions are heading
Bar Counsel’s 2025 article is titled around escalation for a reason. In Johnson v. Dunn, a federal court in Alabama in July 2025 sanctioned lawyers for the state prison commissioner over five fabricated citations, and wrote: “Fabricating legal authority is serious misconduct that demands a serious sanction. It demands substantially greater accountability than the reprimands and modest fines that have become common as courts confront this form of AI misuse. As a practical matter, time is telling us, quickly and loudly, that those sanctions are insufficient deterrents.” The court publicly reprimanded the lawyers, disqualified them from the case, and referred them to the bar. It noted, as a mitigating factor, that their firm had adopted an internal AI policy requiring citation review in 2023, before the incident.
That is the direction. The first-generation sanction was a fine of a few thousand dollars and an apology. The second generation is disqualification, fee-shifting, mandatory education and a bar referral. Massachusetts has already moved from the first to the second in Dastou and Marullo. A lawyer sanctioned in 2026 for hallucinated citations will not be able to argue, as the lawyer in Smith could in early 2024, that the danger was not yet widely known.
The tools, and their limits
Lawyers Weekly reported this week on a new category of software that checks citations in a draft against actual case databases before filing, flagging authorities that do not exist or do not say what the brief claims. Westlaw and Lexis have built similar checks into their own AI products. These tools catch the invented case. They are less reliable at catching the real case with an invented quotation, which was the Dastou problem, and they do nothing about a fabricated doctrine dressed as a jury instruction.
The only reliable check is the one the rules already require: a lawyer who reads every cited case before signing. Bar Counsel’s recommendation to firms is procedural rather than technological. Adopt a written AI policy. Require that every citation in every filing be verified against the source by a human. Train everyone, including staff who may use tools the lawyers do not know about. Document the training, because, as the Alabama court showed, a firm that did those things before the incident is treated differently from one that did not.
What to do if you find a fake citation in a filing
The cases are consistent on this. A lawyer who discovers that a filed document contains fabricated authority must correct it promptly and candidly; Rule 3.3 requires correcting a false statement of law once known. The lawyers in Mata were sanctioned as much for what they did after the court asked as for the original filing. A lawyer on the other side who spots a fake citation should say so in the responsive filing and let the court address it; opposing counsel in the Massachusetts cases were awarded fees for the work of exposing the fabrications. A judge who spots one will issue an order to show cause, and the hearing that follows is where a lawyer’s candor is measured.
Has a Massachusetts lawyer been sanctioned for AI-generated fake citations?
Yes. Smith v. Farwell (Norfolk Superior Court, February 2024, $2,000) and Dastou v. Holmes (Middlesex Superior Court, June 2025, fee forfeiture and mandatory CLE). The Smith lawyer also received a public reprimand from the Board of Bar Overseers in 2025.
Does Massachusetts require lawyers to disclose AI use in court filings?
No. There is no statewide rule or certification requirement. The existing rules on candor, competence, supervision and reasonable inquiry are applied instead.
What rules does a hallucinated citation violate?
Rule 11 (reasonable inquiry before signing), Rule 3.3 (candor toward the tribunal), Rule 1.1 (competence), and, where staff used the tool, Rules 5.1 and 5.3 (supervision). Bar Counsel has also charged Rule 8.4(d).
What is the SJC’s position on generative AI?
Interim guidelines announced in December 2025 govern judges and court staff. No attorney-facing rule has been adopted.
Can a self-represented litigant be sanctioned for fake citations?
Yes. Rule 11 applies to anyone who signs a filing. Courts have imposed fee awards and dismissed appeals over pro se filings built on AI-generated cases.
How do I check whether a citation is real?
Read the case in a primary source, such as the official reporter, Westlaw, Lexis, or a free database like CourtListener, and confirm it says what the brief claims. Citation-checking software can flag nonexistent cases but does not reliably catch fabricated quotations from real ones.
Massachusetts case details from the Office of Bar Counsel’s 2025 article “Two Years of Fake Cases and the Courts Are Ratcheting Up the Sanctions” by Assistant Bar Counsel Heather L. LaVigne, read in full, and from LawSites’ report on Smith v. Farwell, February 2024. Smith v. Farwell, No. 2282-CV-01197 (Norfolk Super. Ct. Feb. 15, 2024); Dastou v. Holmes, No. 2381-CV-02212 (Middlesex Super. Ct. June 25, 2025); Matter of Marullo, 41 Mass. Att’y Disc. R. (2025); Johnson v. Dunn (N.D. Ala. July 23, 2025). SJC interim guidelines from the Chief Justice’s December 2025 address as reported. General information about Massachusetts law, not legal advice.