COLUMN > Ethics & the Law

“[A]ny use of AI requires caution and humility. One of AI’s prominent applications made headlines … for a shortcoming known as ‘hallucination,’ which caused the lawyers using the application to submit briefs with citations to non-existent cases. (Always a bad idea.)”
—Chief Justice John G. Roberts, Jr.11 Chief Justice John G. Roberts Jr., 2023 Year-End Report on the Federal Judiciary at 5-6 (2023).
By Mark J. Fucile
Once upon a time, in an era when paper reigned supreme, there were stories of lawyers who cited cases after only reading the headnotes or similar summaries rather than the opinions themselves.22 See, e.g., Fahner v. Marsh, 1988 WL 5016 (N.D. Ill. Jan. 19, 1988) (unpublished) (lawyer sanctioned under Fed. R. Civ. P. 11 for inadequate legal research when he cited decision for incorrect proposition after only reading about it in a footnote to a CLE publication rather than analyzing the case himself). Those lawyers risked citing cases for the wrong proposition.33 Id. More recently, the case Chief Justice John Roberts was alluding to with his pointed comment—Mata v. Avianca, Inc., 678 F. Supp.3d 443 (S.D.N.Y. 2023)—involved a lawyer who used an artificial intelligence tool to write a brief that included citations to cases that didn’t even exist.44 For a non-technical background discussion of AI tools in law practice that assist with both researching and writing and the associated risk of “hallucinations” in legal citations, see generally Wadsworth v. Walmart, Inc., 348 F.R.D. 489 (D. Wyo. 2025).
Both of these “old” and “new” shortcuts may say more about human nature than the march of technology. With the advent of AI tools that promise both legal research and brief writing, however, the risk from cutting corners is magnified because an entire brief or other critical filing may be stricken. That raises the twin specter of client harm and lawyer sanctions.55 See, e.g., Grant v. City of Long Beach, 96 F.4th 1255 (9th Cir. 2024) (dismissing appeal, in relevant part, for citation to nonexistent case). Depending on the circumstances and the policy involved, AI-related errors may not be covered under at least some malpractice policies. See Danielle Braff, “Got Insurance?” 111 ABA J. 1, 9-10 (Feb./Mar. 2024) (discussing coverage issues from AI-related errors). The growing genre of AI “hallucination” cases has also provided a related illustration of human nature: an unwillingness to “fess up” when lawyers discover mistakes. Failing to acknowledge an initial shortcoming in competence can quickly spiral into a central issue of lawyer credibility with the court. Although judges understandably have little patience for lawyers who took shortcuts by not cite-checking their filings, they have even less tolerance for lawyers who then try to hide their errors.
In this column, we will first briefly note the ethics rule—RPC 3.3(a)(1)—that requires a lawyer to correct a false statement of law the lawyer made to a court. We will then survey the emerging genre of cases where lawyers used AI tools for court filings that included inaccurate—or nonexistent—case citations and the accompanying risk of not promptly correcting the problem once discovered.66 Because this unfortunate genre is still developing, the cases included should not be regarded as a static compilation. Running “hallucination” through a legal research engine will likely continue to yield newer cases.
Before we do, however, four qualifiers are in order.
First, as Supreme Court Chief Justice John Roberts observed, filing a brief without checking the citations is “always a bad idea.” Both the regulatory duty of competence under RPC 1.177 RPC 1.1 provides: “A lawyer shall provide competent representation to a client. Competent representation requires the legal knowledge, skill, thoroughness and preparation reasonably necessary for the representation.” and the civil standard of care88 Washington Pattern Jury Instruction 107.04 on legal malpractice states that “[a]n attorney has a duty to use that degree of skill, care, diligence, and knowledge possessed and used by a reasonable, careful, and prudent attorney in the State of Washington acting in the same or similar circumstances.” put the responsibility for preparing accurate briefs squarely on the lawyer involved. The development of AI tools that both research and write, however, appears to have increased the temptation to take this risky shortcut. Although lawyers have important duties to understand the technology they use in law practice,99 See, e.g., RPC 1.1, cmt. 8 (noting duty to stay current on technology used in law practice). today we will focus on the decidedly more human risk of not owning up to mistaken legal citations once discovered.
Second, we will address lawyers who did not set out to knowingly lie to courts. RPC 3.3(a)(1) prohibits lying outright, and sanctions for intentionally lying to a court are understandably severe.1010 See, e.g., In re Kamb, 177 Wn.2d 851, 305 P.3d 1091 (2013) (lawyer disbarred for altering court order and related misrepresentation); In re Osborne, 187 Wn.2d 188, 386 P.3d 288 (2016) (lawyer disbarred for, in relevant part, filing false declaration).
Third, although we will focus on the ethics rule, many of the examples we will touch on involve court sanctions under Federal Rule of Civil Procedure 11 and state counterparts—under which, using the federal phraseology, a lawyer’s signature on a filing certifies that the “legal contentions are warranted by existing law[.]”1111 For a discussion of Federal Rule of Civil Procedure 11 in the context of nonexistent citations generated by AI tools, see Jessica R. Gunder, “Rule 11 Is No Match for Generative AI,” 27 Stan. Tech. L. Rev. 308 (2024). For a general discussion of Washington Civil Rule 11, see Philip Talmadge, Emmelyn Hart-Biberfeld, and Peter Lohnes, “When Counsel Screws Up: The Imposition and Calculation of Attorney Fees as Sanctions,” 33 Seattle U. L. Rev. 437 (2010). It bears remembering that court sanctions and regulatory discipline are not mutually exclusive and some of the cases we will discuss involve both.1212 See, e.g., People v. Crabill, 2023 WL 8111898 (Colo. Nov. 22, 2023) (unpublished) (lawyer disciplined following court sanctions for failing to check citations to nonexistent cases generated by AI tool used to write brief and then trying to cover-up the error); Park v. Kim, 91 F.4th 610 (2d Cir. 2024) (lawyer referred to disciplinary committee after imposition of sanctions for filing brief that included nonexistent citation generated by AI tool). See also Restatement (Third) of the Law Governing Lawyers § 111, cmt. e (2000) (noting that inadequate legal research can expose lawyers to both regulatory discipline and court sanctions).
Finally, AI is evolving rapidly and, as it does, the ways it will impact both law practice and the legal profession are also in flux. The ABA and state bars around the country—including the WSBA—have undertaken extensive studies of the broader impacts of AI and have or are issuing rolling guidance as the technology evolves.1313 See generally ABA Formal Op. 512 (2024) (addressing AI tools in law practice and compiling authorities nationally). In Washington, the work of the WSBA Legal Technology Task Force is summarized on the WSBA website at: www.wsba.org/connect-serve/committees-boards-other-groups/legal-technology-task-force. We will save those broader considerations for another day. In this column, we will take a small-bore approach to the particular problem of lawyers who don’t acknowledge errors when they discover them in briefs that include inaccurate citations generated by AI tools.1414 Some of the cases in the AI “hallucination” genre involve pro se litigants. See, e.g., Kruse v. Karlen, 692 S.W.3d 43 (Mo. App. 2024); Al-Hamim v. Star Hearthstone, LLC, 564 P.3d 1117, 2024 WL 5230126 (Colo. App. 2024). In this column, we focus on lawyers. That said, lawyers who “ghostwrite” briefs for pro se litigants should carefully verify the accuracy of the legal citations included. See generally WSBA Advisory Op. 202002 (2020) (lawyer ghostwriting for pro se litigants); ABA Formal Op. 07-446 (2007) (same).
RPC 3.3(a)(1)
RPC 3.3(a)(1) is simple and direct:
A lawyer shall not knowingly:
(1) make a false statement of fact or law to a tribunal or fail to correct a false statement of material fact or law previously made to the tribunal by the lawyer.1515 Correcting statements of fact was addressed in this space as “Reckoning: The Duty to Correct under RPC 3.3,” in the December 2024/January 2025 issue. RPC 3.3(a)(3) addresses the associated principle that a lawyer must “disclose to the tribunal legal authority in the controlling jurisdiction known to the lawyer to be directly adverse to the position of the client and not disclosed by the opposing party[.]”
Comment 4 to RPC 3.3 states the reason for the rule:
Legal argument based on a knowingly false representation of law constitutes dishonesty toward the tribunal.
Washington’s rule is patterned on its ABA Model Rule counterpart.1616 Robert H. Aronson, “An Overview of the Law of Professional Responsibility: The Rules of Professional Conduct Annotated and Analyzed,” 61 Wash. L. Rev. 823, 863 (1986) (discussing Washington’s adoption of the RPCs based on the ABA Model Rules in 1985). Both trace their lineage to the ABA Canon 22—“Candor and Fairness”—adopted by the ABA in 1908.
As noted earlier, today we will focus on the failure to correct once an inaccuracy has been discovered, and the cases in the “hallucination” genre thus far tilt in that direction. Absent a technological fix, however, it is conceivable that with the growing awareness of that shortcoming a lawyer who uses an AI tool to both research and write a brief without independently checking the accuracy of the resulting citations could at some point be accused of knowingly making a false statement of law from the outset—perhaps in a disciplinary setting but more likely in a sanctions context. RPC 1.0A(f), for example, notes that knowledge can be inferred from the circumstances. Washington Civil Rule 11(a)(2) and Federal Rule of Civil Procedure 11(b)(2), in turn, both frame a lawyer’s certification of the legal authority cited in filings as being based on “an inquiry reasonable under the circumstances.”1717 On a related note, some courts around the country by local rule or general order require disclosure of the use of AI tools to varying degrees for filings. For a convenient tracker of these kinds of rules and orders, see www.ropesgray.com/en/sites/artificial-intelligence-court-order-tracker.
Hallucination Cases
Nationally, “hallucination” cases are a comparatively diverse lot. Some involve experienced partners.1818 See, e.g., Mata v. Avianca, Inc., supra, 678 F. Supp.3d 443. Others involve junior associates.1919 See, e.g., People v. Crabill, supra, 2023 WL 8111898. Some are civil cases.2020 See, e.g., Will of Samuel, 206 N.Y.S.3d 888 (N.Y. Surrogate’s Ct. 2024). Others come from criminal practice.2121 See, e.g., United States v. Cohen, 724 F. Supp.3d 251 (S.D.N.Y. 2024). Some involve large law firms.2222 See, e.g., Wadsworth v. Walmart, supra, 2025 WL 608073. Others involve solos.2323 See, e.g., Park v. Kim, supra, 91 F.4th 610. Some involve lawyers using consumer products.2424 See, e.g., Mata v. Avianca, supra, 678 F. Supp.3d 443. Others involve products tailored to law practice.2525 See, e.g., Gauthier v. Goodyear Tire & Rubber Co., 2024 WL 4882651 (E.D. Tex. Nov. 25, 2024) (unpublished). A few involve lawyers who said they understood how the AI tool worked.2626 Park v. Kim, supra, 91 F.4th at 614. Others claimed ignorance.2727 See, e.g., Mata v. Avianca, Inc., supra, 678 F. Supp.3d at 456. Although a few involve discrete citations in briefs otherwise prepared by lawyers,2828 See, e.g., United States v. Cohen, supra, 724 F. Supp.3d at 254-55, 257-260. many involve multiple citations to nonexistent cases in filings prepared largely by AI tools.2929 See, e.g., Mata v. Avianca, supra, 678 F. Supp.3d 443; People v. Crabill, supra, 2023 WL 8111898; Smith v. Farwell, 2024 WL 4002576 (Mass. Super. Feb. 15, 2024) (unpublished). What they all share is Chief Justice Roberts’ pungent observation: The lawyers didn’t check their work.
In almost all instances, either opposing counsel or the court discovered the error. What happened next often dictated the outcome for the lawyers.
Some refused—at least initially—to acknowledge the source of the error. A case from Colorado involving a young associate provides a telling example:
When the judge expressed concerns about the accuracy of the cases, Crabill falsely attributed the mistakes to a legal intern. Six days after the hearing, Crabill filed an affidavit with the court, explaining that he used ChatGPT when he drafted the motion.3030 People v. Crabill, supra, 2023 WL 8111898 at *1.
The lawyer was disciplined under Colorado’s version of RPC 3.3(a)(1) for failing to disclose the error once it surfaced—along with Colorado RPC 1.1 on competence and RPC 8.4(c) for engaging in dishonest conduct.3131 Id. Media reports noted that the lawyer was later fired from his law firm.3232 Pranshu Verma and Will Oremus, “These Lawyers Used ChatGPT to Save Time. They Got Fired and Fined,” Washington Post, Nov. 16, 2023 (available at: www.washingtonpost.com/technology/2023/11/16/chatgpt-lawyer-fired-ai/#).
By contrast, others immediately accepted responsibility and apologized.
A case from New York involving a more seasoned lawyer provides an equally telling example:
The Court also credits Schwartz’s representations—that the inclusion of the cases in his motion was an “honest” and “unfortunate mistake,” … that he “had no intention to deceive the Court,” … and that he would have withdrawn the citations immediately if given the opportunity … and does not doubt the genuineness of his apologies and acceptance of responsibility.3333 United States v. Cohen, supra, 724 F. Supp.3d at 258-59.
In light of the forthright apology, the court did not impose sanctions.3434 Id. at 260.
This is not to suggest that a quick apology will cure all. A Massachusetts court, for example, noted its appreciation for a lawyer’s ready acceptance of responsibility and apology, but still fined the lawyer $2,000.3535 Smith v. Farwell, supra, 2024 WL 4002576. See also Kohls v. Ellison, 2025 WL 66514 at *4-*5 (D. Minn. Jan. 10, 2025) (unpublished) (noting a quick and sincere apology when a “hallucination” was discovered in an expert’s declaration but still striking the declaration). Moreover, an insincere apology can heighten rather than lessen the risk of sanctions.3636 See, e.g., United States v. Hayes, 763 F. Supp. 3d 1045, 1066 (E.D. Cal. 2025). (“The Court finds this response inadequate and not credible.”); University of Washington v. Singh, 2024 WL 4315178 at *5 (W.D. Wash. Sept. 26, 2004) (unpublished) (“[T]he court finds it strains credulity that the originally proffered case could have been mistaken for the ‘substitute’ case.”).
Nonetheless, as the old saw attributed to Will Rogers counsels: “If you find yourself in a hole, stop digging.”3737 See www.parks.ca.gov/?page_id=23998 (California State Parks/Will Rogers State Park website). Lawyers have a clear responsibility for their filings under RPC 3.3(a)(1), applicable court rules such as CR 11 and its federal counterpart, and the standard of care. Failure to meet those duties can have serious consequences for both the lawyers and their clients.3838 Where potential client harm is involved, conflicts can follow under RPC 1.7(a)(2) between the interest of the client and the interest of the lawyer. Depending on the circumstances, some of those conflicts may be waivable, but others may not (and, therefore, would require the lawyer’s withdrawal under RPC 1.16(a)(1)). Those consequences are ordinarily stark enough when a lawyer has failed to check the accuracy of the legal citations in a filing and they are incorrect, or even worse, nonexistent. To compound those consequences by not being completely honest with the court is to invite turning a “hallucination” into a very unpleasant reality of potential court sanctions, regulatory discipline and, depending on the circumstances, civil claims. Further, once lost, credibility with the court can be difficult to restore.3939 See, e.g., Lee v. Delta Air Lines, Inc., 2024 WL 1230263 at *3 (E.D.N.Y. Mar. 22, 2024) (unpublished) (“[T]he Court maintains serious concern that at least one of Plaintiff’s cited cases is non-existent and may have been a hallucinated product of generative artificial intelligence, particularly given Plaintiff’s recent history of similar conduct before the Second Circuit.”). In short, while it is always a bad idea to file a brief without cite-checking it as Chief Justice Roberts noted, not fessing up to the court when that failing inevitably surfaces is even worse.
NOTES
1. Chief Justice John G. Roberts Jr., 2023 Year-End Report on the Federal Judiciary at 5-6 (2023).
2. See, e.g., Fahner v. Marsh, 1988 WL 5016 (N.D. Ill. Jan. 19, 1988) (unpublished) (lawyer sanctioned under Fed. R. Civ. P. 11 for inadequate legal research when he cited decision for incorrect proposition after only reading about it in a footnote to a CLE publication rather than analyzing the case himself).
3. Id.
4. For a non-technical background discussion of AI tools in law practice that assist with both researching and writing and the associated risk of “hallucinations” in legal citations, see generally Wadsworth v. Walmart, Inc., 348 F.R.D. 489 (D. Wyo. 2025).
5. See, e.g., Grant v. City of Long Beach, 96 F.4th 1255 (9th Cir. 2024) (dismissing appeal, in relevant part, for citation to nonexistent case). Depending on the circumstances and the policy involved, AI-related errors may not be covered under at least some malpractice policies. See Danielle Braff, “Got Insurance?” 111 ABA J. 1, 9-10 (Feb./Mar. 2024) (discussing coverage issues from AI-related errors).
6. Because this unfortunate genre is still developing, the cases included should not be regarded as a static compilation. Running “hallucination” through a legal research engine will likely continue to yield newer cases.
7. RPC 1.1 provides: “A lawyer shall provide competent representation to a client. Competent representation requires the legal knowledge, skill, thoroughness and preparation reasonably necessary for the representation.”
8. Washington Pattern Jury Instruction 107.04 on legal malpractice states that “[a]n attorney has a duty to use that degree of skill, care, diligence, and knowledge possessed and used by a reasonable, careful, and prudent attorney in the State of Washington acting in the same or similar circumstances.”
9. See, e.g., RPC 1.1, cmt. 8 (noting duty to stay current on technology used in law practice).
10. See, e.g., In re Kamb, 177 Wn.2d 851, 305 P.3d 1091 (2013) (lawyer disbarred for altering court order and related misrepresentation); In re Osborne, 187 Wn.2d 188, 386 P.3d 288 (2016) (lawyer disbarred for, in relevant part, filing false declaration).
11. For a discussion of Federal Rule of Civil Procedure 11 in the context of nonexistent citations generated by AI tools, see Jessica R. Gunder, “Rule 11 Is No Match for Generative AI,” 27 Stan. Tech. L. Rev. 308 (2024). For a general discussion of Washington Civil Rule 11, see Philip Talmadge, Emmelyn Hart-Biberfeld, and Peter Lohnes, “When Counsel Screws Up: The Imposition and Calculation of Attorney Fees as Sanctions,” 33 Seattle U. L. Rev. 437 (2010).
12. See, e.g., People v. Crabill, 2023 WL 8111898 (Colo. Nov. 22, 2023) (unpublished) (lawyer disciplined following court sanctions for failing to check citations to nonexistent cases generated by AI tool used to write brief and then trying to cover-up the error); Park v. Kim, 91 F.4th 610 (2d Cir. 2024) (lawyer referred to disciplinary committee after imposition of sanctions for filing brief that included nonexistent citation generated by AI tool). See also Restatement (Third) of the Law Governing Lawyers § 111, cmt. e (2000) (noting that inadequate legal research can expose lawyers to both regulatory discipline and court sanctions).
13. See generally ABA Formal Op. 512 (2024) (addressing AI tools in law practice and compiling authorities nationally). In Washington, the work of the WSBA Legal Technology Task Force is summarized on the WSBA website at: www.wsba.org/connect-serve/committees-boards-other-groups/legal-technology-task-force.
14. Some of the cases in the AI “hallucination” genre involve pro se litigants. See, e.g., Kruse v. Karlen, 692 S.W.3d 43 (Mo. App. 2024); Al-Hamim v. Star Hearthstone, LLC, 564 P.3d 1117, 2024 WL 5230126 (Colo. App. 2024). In this column, we focus on lawyers. That said, lawyers who “ghostwrite” briefs for pro se litigants should carefully verify the accuracy of the legal citations included. See generally WSBA Advisory Op. 202002 (2020) (lawyer ghostwriting for pro se litigants); ABA Formal Op. 07-446 (2007) (same).
15. Correcting statements of fact was addressed in this space as “Reckoning: The Duty to Correct under RPC 3.3,” in the December 2024/January 2025 issue. RPC 3.3(a)(3) addresses the associated principle that a lawyer must “disclose to the tribunal legal authority in the controlling jurisdiction known to the lawyer to be directly adverse to the position of the client and not disclosed by the opposing party[.]”
16. Robert H. Aronson, “An Overview of the Law of Professional Responsibility: The Rules of Professional Conduct Annotated and Analyzed,” 61 Wash. L. Rev. 823, 863 (1986) (discussing Washington’s adoption of the RPCs based on the ABA Model Rules in 1985).
17. On a related note, some courts around the country by local rule or general order require disclosure of the use of AI tools to varying degrees for filings. For a convenient tracker of these kinds of rules and orders, see www.ropesgray.com/en/sites/artificial-intelligence-court-order-tracker.
18. See, e.g., Mata v. Avianca, Inc., supra, 678 F. Supp.3d 443.
19. See, e.g., People v. Crabill, supra, 2023 WL 8111898.
20. See, e.g., Will of Samuel, 206 N.Y.S.3d 888 (N.Y. Surrogate’s Ct. 2024).
21. See, e.g., United States v. Cohen, 724 F. Supp.3d 251 (S.D.N.Y. 2024).
22. See, e.g., Wadsworth v. Walmart, supra, 2025 WL 608073.
23. See, e.g., Park v. Kim, supra, 91 F.4th 610.
24. See, e.g., Mata v. Avianca, supra, 678 F. Supp.3d 443.
25. See, e.g., Gauthier v. Goodyear Tire & Rubber Co., 2024 WL 4882651 (E.D. Tex. Nov. 25, 2024) (unpublished).
26. Park v. Kim, supra, 91 F.4th at 614.
27. See, e.g., Mata v. Avianca, Inc., supra, 678 F. Supp.3d at 456.
28. See, e.g., United States v. Cohen, supra, 724 F. Supp.3d at 254-55, 257-260.
29. See, e.g., Mata v. Avianca, supra, 678 F. Supp.3d 443; People v. Crabill, supra, 2023 WL 8111898; Smith v. Farwell, 2024 WL 4002576 (Mass. Super. Feb. 15, 2024) (unpublished).
30. People v. Crabill, supra, 2023 WL 8111898 at *1.
31. Id.
32. Pranshu Verma and Will Oremus, “These Lawyers Used ChatGPT to Save Time. They Got Fired and Fined,” Washington Post, Nov. 16, 2023 (available at: www.washingtonpost.com/technology/2023/11/16/chatgpt-lawyer-fired-ai/#).
33. United States v. Cohen, supra, 724 F. Supp.3d at 258-59.
34. Id. at 260.
35. Smith v. Farwell, supra, 2024 WL 4002576. See also Kohls v. Ellison, 2025 WL 66514 at *4-*5 (D. Minn. Jan. 10, 2025) (unpublished) (noting a quick and sincere apology when a “hallucination” was discovered in an expert’s declaration but still striking the declaration).
36. See, e.g., United States v. Hayes, 763 F. Supp. 3d 1045, 1066 (E.D. Cal. 2025). (“The Court finds this response inadequate and not credible.”); University of Washington v. Singh, 2024 WL 4315178 at *5 (W.D. Wash. Sept. 26, 2004) (unpublished) (“[T]he court finds it strains credulity that the originally proffered case could have been mistaken for the ‘substitute’ case.”).
37. See www.parks.ca.gov/?page_id=23998 (California State Parks/Will Rogers State Park website).
38. Where potential client harm is involved, conflicts can follow under RPC 1.7(a)(2) between the interest of the client and the interest of the lawyer. Depending on the circumstances, some of those conflicts may be waivable, but others may not (and, therefore, would require the lawyer’s withdrawal under RPC 1.16(a)(1)).
39. See, e.g., Lee v. Delta Air Lines, Inc., 2024 WL 1230263 at *3 (E.D.N.Y. Mar. 22, 2024) (unpublished) (“[T]he Court maintains serious concern that at least one of Plaintiff’s cited cases is non-existent and may have been a hallucinated product of generative artificial intelligence, particularly given Plaintiff’s recent history of similar conduct before the Second Circuit.”).

