特拉华衡平法院裁定律师用 GenAI 编造引文违反 Rule 11,签字律师个人被罚 2500 美元
Kevin Leiske v. Robert Gregory Kidd
特拉华州衡平法院 11 月 13 日就 Kevin Leiske 诉 Robert Gregory Kidd 案作出裁定,原告首席律师在答复简报中使用未获律所批准的 Perplexity Enterprise 起草内容,带入虚构引文和错误判例描述,违反 Rule 11(b)(2)。
特拉华州衡平法院认定律师用 GenAI 生成虚构引文违反 Rule 11,签字律师个人罚 2500 美元,律所因已有 GenAI 政策与培训免于连带责任。
译文尚不完整,完整内容请切换到原文。
跳至
支持 FLP
CourtListener 是 Free Law Project 的项目,该组织是联邦认可的 501(c)(3) 非营利机构。会员可帮助支持我们的工作,并获得功能的特别访问权限。
请今天就成为会员。
2026年11月13日
下载 PDF
添加备注
Kevin Leiske 诉 Robert Gregory Kidd
特拉华州衡平法院
合并意见书
COURT OF CHANCERY
OF THE
STATE OF DELAWARE
LORI W. WILL LEONARD L. WILLIAMS JUSTICE CENTER
副大法官 500 N. KING STREET, SUITE 11400
WILMINGTON, DELAWARE 19801-3734
2026年11月13日
Richard P. Rollo, Esquire Margaret M. DiBianca, Esquire
Travis S. Hunter, Esquire DiBianca Law, LLC
Richards, Layton & Finger, P.A. 1201 North Orange Street, Suite 504
920 North King Street Wilmington, Delaware 19801
Wilmington, Delaware 19801
RE: Kevin Leiske et al. v. Robert Gregory Kidd et al.,
C.A. No. 2025-0426-CDW (LWW)
尊敬的律师:
This letter resolves an Order to Show Cause concerning the misuse of
生成式人工智能(“GenAI”)。原告向本法院提交的一份书状中使用了捏造的
引文,并对法律依据作了错误描述。
在被告指出这些错误后,原告律师承认
使用了 GenAI 工具却未核实其输出,并提交了一份更正后的书状,
删除了引号,但保留了不准确的法律主张。我命令
签署该书状的合伙人和其律师事务所说明理由,解释为何不应
施加制裁。
The use of GenAI in legal practice is neither forbidden nor inherently
问题。执业者越来越多地使用这些工具,它们带来了
显著的好处和效率提升的机会。但律师必须按照
C.A. No. 2025-0426-CDW (LWW)
2026年11月13日
第 2 页,共 13 页
衡平法院规则第 11 条和《特拉华州律师职业行为规则》
所要求的谨慎来使用它们。这一义务包括确保提交给法院的每一项
法律依据和引证都准确无误。
That care was lacking here. Counsel failed to verify the accuracy of the
答辩书状,尽管其律师事务所制定了全面的 GenAI 政策并开展了强制性
伦理培训。鉴于该事务所为防止 GenAI 滥用所做的值得称赞的努力,我
决定不对其施加制裁。该律师——此后已承担全部责任并
表达了真诚的悔意——依据规则第 11 条被处以 2,500 美元罚款,
须由其个人向衡平法院支付。
一、 背景
On January 22, 2026, the plaintiffs filed an answering brief opposing the
被告对衡平法院治安法官关于预付款问题的最终报告提出的异议。1
五天后,被告的答辩书状指出,该答辩书状
包含捏造的引文、不准确的引证以及幻觉产生的法律
主张。2
On January 28, plaintiffs’ lead counsel—the signatory on the brief—filed a
一封信函,承认曾使用 GenAI 工具修改书状,并
1
原告针对被告异议的答辩书状(案卷 62)。
2
关于进一步支持被告对治安法官 2026 年 1 月 2 日报告的异议的答辩书状
(案卷 65)12-13 页注 4-6,16 页注 7。
C.A. No. 2025-0426-CDW (LWW)
2026年11月13日
第 3 页,共 13 页
提交了一份更正后的答辩书状。3 该信函将未能发现
fabricated quotations and citation errors to an oversight during a paralegal’s cite-
check.4 Although the corrected brief removed quotation marks around the
nonexistent quotes, it retained the inaccurate descriptions of the underlying cases.5
After the defendants’ counsel filed a letter detailing these remaining
misrepresentations, the plaintiffs’ lead counsel responded on January 29 by faulting
opposing counsel for failing to “meet and confer” before alerting the court and
insisting that the cited authorities “are real” and “support the arguments advanced.”6
On February 2, I overruled the defendants’ exceptions and affirmed the
Magistrate’s final report on the merits, while retaining jurisdiction to address the
GenAI errors in the plaintiffs’ brief.7 On July 1, after further merits-based
proceedings, I entered an Order to Show Cause directing the plaintiffs’ lead counsel
and his law firm to explain why sanctions should not issue under Rule 11(c) and this
3
Letter Providing Corrected Answering Br. in Opp’n to Defs.’ Exceptions (Dkt. 68); Pls.’
Corrected Answering Br. in Opp’n to Defs.’ Exceptions (Dkt. 69) (“Pls.’ Corrected
Answering Br.”).
4
Letter Providing Corrected Answering Br. in Opp’n to Defs.’ Exceptions (Dkt. 68) 2.
5
See Letter in Resp. to Pls.’ Letter Regarding GenAI Errors (Dkt. 70) 1-2 (asserting that
the corrected brief “merely fixes citation errors as if they are typographical only” and “still
mischaracterizes some of the Courts’ analyses”).
6
Letter in Resp. to Defs.’ Jan. 28 Letter (Dkt. 71) 1-2.
7
Letter Op. Resolving Exceptions (Dkt. 74) 7 n.40.
C.A. No. 2025-0426-CDW (LWW)
November 13, 2026
Page 4 of 13
court’s inherent authority.8 Richard Rollo, Esq., on his own behalf, and Paul N.
Heath, Esq., on behalf of Richards, Layton & Finger, P.A. (“RLF”), filed detailed
affidavits in response to that Order.9 Having reviewed those affidavits, I conclude
that a separate evidentiary hearing is unnecessary and that the matter can be resolved
on the papers.10
II. ANALYSIS
Under Rule 11(b)(2), an attorney presenting “a pleading, written motion, or
other paper” to the court certifies that “to the best of the person’s knowledge,
information, and belief, formed after an inquiry reasonable under the
circumstances,” the “legal contentions” made “are warranted by existing law.”11
“An attorney’s signature on a court filing is not a procedural formality; it is a
certification to the tribunal.”12 A fake quotation or a fabricated legal synthesis
8
Letter Order to Show Cause Why Sanctions Should Not Issue (Dkt. 102) (“Rule 11 Letter
Order”) 6-8 (specifying what the affidavits from counsel and the firm must include).
9
Aff. of R. Rollo, Esq. (Dkt. 103) (“Rollo Aff.”); Aff. of P. Heath, Esq. (Dkt. 103) (“Heath
Aff.”).
10
See Hill v. Workday, Inc., 2026 WL 1146289, at *4 (N.D. Cal. Apr. 28, 2026) (observing
that a separate sanctions hearing is not required when counsel gets a chance to respond in
writing) (citing Pac. Harbor Cap., Inc. v. Carnival Air Lines, Inc., 210 F.3d 1112, 1118
(9th Cir. 2000)).
11
Ct. Ch. R. 11(b)(2).
12
Rule 11 Letter Order 4 (citing Crumplar v. Superior Ct. ex rel. New Castle Cnty., 56 A.3d 1000, 1005-06 (Del. 2012)).
C.A. No. 2025-0426-CDW (LWW)
November 13, 2026
Page 5 of 13
generated by an algorithm is not “existing law,” and presenting unverified GenAI
output to a court violates Rule 11(b)(2).13
A. Law Firm Responsibility Under Rule 11
Rule 11(c)(1) mandates that “[a]bsent exceptional circumstances, a law firm
must be held jointly responsible for violations committed by its partners, associates,
or employees.”14 In determining whether a law firm should be relieved of joint
responsibility for a lawyer’s Rule 11 violation, professional conduct standards
governing law firm management provide useful guidance. Under Delaware
Lawyers’ Rule of Professional Conduct 5.1(a), law firm leadership must make
reasonable efforts to ensure that the firm has measures in effect giving reasonable
assurance that all lawyers conform to those rules.15 Consistent with that principle,
ABA Formal Opinion 512 instructs “[m]anagerial lawyers” to “establish clear
policies regarding the law firm’s permissible use of G[en]AI” and “supervisory
13
Ct. Ch. R. 11(b)(2); see Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 461 (S.D.N.Y. 2023)
(“A fake opinion is not ‘existing law’ . . . .”); see also Lexos Media IP, LLC v.
Overstock.com, Inc., 2026 WL 265581, at *10 (D. Kan. Feb. 2, 2026) (“There is no
question that citing to a nonexistent case, attributing a nonexistent quotation to an existing
case, and misstating the law violates Rule 11(b).”).
14
Ct. Ch. R. 11(c)(1).
15
Del. Lawyers’ Rules of Prof. Conduct R. 5.1(a) (“A partner in a law firm, and a lawyer
who individually or together with other lawyers possesses comparable managerial
authority in a law firm, shall make reasonable efforts to ensure that the firm has in effect
measures giving reasonable assurance that all lawyers in the firm conform to the Rules of
Professional Conduct.”).
C.A. No. 2025-0426-CDW (LWW)
November 13, 2026
Page 6 of 13
lawyers” to “make reasonable efforts to ensure that the firm’s lawyers and
nonlawyers comply with their professional obligations when using G[en]AI tools.”16
RLF has demonstrated exceptional circumstances relieving it of joint
responsibility. Its response to the emergence of GenAI reflects substantial
institutional efforts to prevent the very problem at issue. RLF barred the use of
unapproved GenAI tools in March 2023.17 In 2024, the firm formed a working group
of directors and professional staff to evaluate GenAI tools and study ethical
guidance, including ABA Formal Opinion 512.18 In July 2025, RLF’s Board of
Directors adopted a comprehensive Generative Artificial Intelligence Policy.19 The
policy requires an experienced lawyer to review GenAI output for accuracy and
completeness and prohibits its use as a substitute for independent judgment or legal
analysis.20 RLF also developed an internal training program addressing
hallucinations, ABA Formal Opinion 512, and professional obligations concerning
competence, confidentiality, meritorious claims, candor, and supervision.21 The
16
ABA Comm. on Ethics & Prof. Resp., Formal Op. 512, at 10 (2024); see Lexos Media,
2026 WL 265581, at *16.
17
Heath Aff. ¶ 8.
18
Id. ¶¶ 9-14.
19
Id. ¶ 15.
20
Id. ¶¶ 15-16.
21
Id. ¶¶ 23, 25-28.
C.A. No. 2025-0426-CDW (LWW)
November 13, 2026
Page 7 of 13
program was accredited for one hour of Enhanced Ethics credit by Delaware’s
Commission on Continuing Legal Education and completed by over 200 RLF
employees by January 2026.22
In short, RLF did the right thing. It warned of GenAI risks, established clear
verification policies, and trained its personnel through a substantial, multi-layered
approach. RLF acted reasonably to prevent GenAI misuse and should not be
sanctioned for an individual lawyer bypassing those controls.23 These safeguards
在问题简报提交之前就已到位。因此,RLF 免于
本次程序中的制裁。
B. Individual Responsibility Under Rule 11
Rule 11(c)(1) authorizes the court to impose an appropriate sanction on “any
律师……违反了[规则 11(b)]或应对违规负责。”24
规则的认证要求强化了律师根据《特拉华州律师职业行为规则》3.3(a)(1)和
3.3 对法庭负有的坦诚道德义务。
22
同上,第 24、29 段。
23
参见 Johnson v. Dunn, 792 F. Supp. 3d 1241, 1260-62 (N.D. Ala. 2025)(拒绝
根据法院固有权力对一家律师事务所实施制裁,因为该所已警告
律师注意幻觉风险,要求使用生成式人工智能须经许可并进行独立核实,
并在发现不当行为后加强了保障措施,认定该所在
一名合伙人滥用生成式人工智能时“已合理努力防止此类不当行为”)。
24
Ct. Ch. R. 11(c)(1)。
C.A. No. 2025-0426-CDW (LWW)
2026 年 11 月 13 日
第 8 页,共 13 页
根据规则 5.3 监督非律师助理的义务。25 正如美国律师协会正式
第 512 号意见所强调的,律师不得将专业判断委托给
生成式人工智能工具,并且仍须负责在提交前核实所有工作成果。26
The affidavit of plaintiffs’ lead counsel candidly explains how the errors
发生了。作为 RLF 人工智能小组委员会主席,他当时正牵头对
Perplexity Enterprise 进行试点评估,该工具尚未获准在全所通用。27 他
使用该工具帮助起草和精简答辩简报中的论点。这一
过程在文本中引入了捏造的引文、有缺陷的引证以及歪曲的
案件判决。28 随后他将引证核查委托给一名初级律师
和一名律师助理,并在未亲自核实权威依据、也未在审查后
检查引证和引文的情况下提交了简报。29
Three points warrant emphasis.
First, the duty of candor to the court and the obligation to conduct a reasonable
规则 11 项下的调查是个人的且不可委托的。30 尽管从业者
25
Del. Lawyers’ Rules of Prof. Conduct R. 3.3(a)(1), 5.3。
26
ABA Comm. on Ethics & Pro. Resp., Formal Op. 512, at 3-4, 10 (2024)。
27
Rollo Aff. ¶¶ 9, 25;另见 Heath Aff. ¶¶ 17, 19, 33-34。
28
Rollo Aff. ¶¶ 10-13。
29
同上,第 14-16 段。
30
参见 Pavelic & LeFlore v. Marvel Ent. Grp., 493 U.S. 120, 126 (1989)(“规则 11 的
整体目的是让签署人个人认识到其个人、不可委托的
C.A. No. 2025-0426-CDW (LWW)
2026 年 11 月 13 日
第 9 页,共 13 页
通常且适当地依赖熟练的律师助理和初级律师进行引证
核查,但签署简报的资深律师在未经核实的生成式人工智能输出
被提交给法院时不能推卸责任。31 当律师明知
将生成式人工智能生成的文本引入草稿时,签署人必须确保
已审查所依据的权威依据,并确认其支持引用它们所主张的
命题。
Second, it is no excuse that the cases cited are real decisions or that the
总体法律命题可以说与一般法律原则相符。生成式人工智能工具不仅
会虚构案件标题。它们还会捏造听起来合理的
引文,或曲解真实判决,使其支持从未涉及的命题。32 现有法律可能总体上支持一方的最终立场
addressed.32 That existing law may generally support a party’s ultimate position
responsibility.”); Lexos Media, 2026 WL 265581, at *9 (noting that the Rule 11 duty to
conduct a reasonable inquiry into existing law is “nondelegable”).
31
See Hill, 2026 WL 1146289, at *3, *8 (“The duty to check every paper filed in court for
accuracy and veracity is nondelegable to other attorneys or to technology.” (citation
omitted)); In re Kenney, 422 So. 3d 905, 913, 920-21 (La. Ct. App. 2025) (affirming
sanctions against counsel who relied on paralegals to check citations but failed to
independently verify AI-generated authorities against actual judicial opinions); see also
Noland v. Land of the Free, L.P., 336 Cal. Rptr. 3d 897, 901 (Ct. App. 2025) (“[N]o brief,
pleading, motion, or any other paper filed in any court should contain any citations—
whether provided by generative AI or any other source—that the attorney responsible for
submitting the pleading has not personally read and verified.”).
32
See, e.g., Lexos Media, 2026 WL 265581, at *2-5 (detailing GenAI hallucinations
involving fabricated quotations attributed to actual decisions and mischaracterizations of
their holdings); Hill, 2026 WL 1146289, at *2 (addressing a brief that cited a nonexistent
case and misrepresented the holding of an actual decision); Wilson v. Kipp Texas, Inc.,
C.A. No. 2025-0426-CDW (LWW)
November 13, 2026
Page 10 of 13
does not excuse the submission of fabricated quotations or mischaracterized
holdings.33 If GenAI errors are detected, counsel has an affirmative duty to review
the underlying authorities and correct any inaccurate quotations or characterizations.
It is not enough to remove quotation marks and leave a fabricated legal synthesis in
place.34
Third, as explained in the Order to Show Cause, faulting opposing counsel for
failing to “meet and confer” before alerting the court to GenAI errors misstates the
purpose of that requirement.35 There is nothing to negotiate when false citations or
2025 WL 3030444, at *3 (N.D. Tex. Oct. 29, 2025) (sanctioning counsel for presenting
GenAI-fabricated statements as verbatim quotations in a court filing).
33
See Johnson, 792 F. Supp. 3d at 1262 (rejecting an argument that sanctions should be
minimal because “actual authorities stand for the proposition that the bogus authorities
were offered to support,” and noting that “any sanctions discount on this basis would
amplify the siren call of unverified AI for lawyers who are already confident in their legal
conclusion”); Lexos Media, 2026 WL 265581, at *3-5, *13-14 (sanctioning counsel where
a brief attributed nonexistent quotations and holdings to real precedent and rejecting
counsel’s attempt to downplay the errors by claiming that the brief was overwhelmingly
human-drafted).
34
See Mata, 678 F. Supp. 3d at 461, 464 (noting that Rule 11 obligations continue when
“later advocating” a prior filing and sanctioning counsel who continued to defend
fabricated authorities after opposing counsel flagged them); Hill, 2026 WL 1146289, at *3
(“Upon detection of fabricated or erroneous citations, an attorney has an affirmative,
personal, non-delegable duty to review all citations for accuracy and validity and to correct
them . . . .” (citing Kenney, 422 So. 3d at 915-16, 920-21)).
35
Rule 11 Letter Order 5-6.
C.A. No. 2025-0426-CDW (LWW)
November 13, 2026
Page 11 of 13
fabricated quotations are presented to a tribunal. Opposing counsel was fully
有理由将此类错误直接提请法院注意。36
C. Sanctions
A Rule 11 sanction “must be limited to what suffices to deter repetition of the
“由其他处境类似者实施的行为或类似行为。”37 此类
制裁的目的是维护司法权威并遏制疏忽——而非惩罚或
补偿诉讼当事人。38
In setting an appropriate sanction, I consider the specific circumstances of this
案件。一方面,原告首席律师最初将责任推给一名律师助理,
并且在删除引号后,仍保留了不准确的表述
36
2026年8月4日,被告律师提交了一封信函,请求准许提交
账单记录以及有关原告律师时间记录的信息。来自
M. DiBianca 的信函(Dkt. 104)。该请求被驳回。由于本次 Rule 11 调查是
依职权主动启动,旨在处理影响法院的不当行为,而非基于制裁动议,
因此根据 Rule 11 无法判给律师费。参见 Ct. Ch. R. 11(c)(4);Lexos
Media, 2026 WL 265581, at *7 n.63;Mata, 678 F. Supp. 3d at 465-66。由于
原告律师已在宣誓后承认违反 Rule 11,且我采信针对说明理由令
提交的宣誓陈述书,邀请对方
律师就账单记录中所谓的差异提出异议将毫无意义。
37
Ct. Ch. R. 11(c)(4);accord Fed. R. Civ. P. 11(c)(4);另见 Cooter & Gell v. Hartmarx
Corp., 496 U.S. 384, 393 (1990)(将遏制确定为 Rule 11 的核心目的)。
38
参见 Mata, 678 F. Supp. 3d at 466(“律师可能被要求支付罚款,或者,用 Rule 11 的
措辞来说,支付‘处罚金’,以促进遏制之利益,而非作为惩罚
或补偿。”(引用 Universitas Educ., LLC v. Nova Grp., Inc., 784 F.3d 99, 103-04
(2d Cir. 2015)));Johnson, 792 F. Supp. 3d at 1265-66(强调在依法院固有权力
制定制裁时应注重遏制和
维护司法权威)。
C.A. No. 2025-0426-CDW (LWW)
2026年11月13日
第 12 页,共 13 页
法律依据。另一方面,律师在其宣誓陈述书中完全接受
对提交文件负责,否认任何暗示他人应受责备的说法,
并表达了真诚的悔意。39 他多年来在本法院执业,
表现卓越。40 而且,由于他已完成了多门关于
生成式人工智能伦理的 CLE 课程,并领导内部培训工作,下令追加强制性的 CLE
将是多余的。41
Consistent with the range of monetary penalties imposed by courts in similar
在依职权启动的 Rule 11 程序中,原告首席律师被处以
2,500 美元的制裁。42 鉴于 Rule 11 核实的不可委托性质
39
Rollo 宣誓书 ¶¶ 3-5, 22, 30。
40
同上 ¶¶ 1, 30。
41
同上 ¶¶ 24-26;参见 Mata, 678 F. Supp. 3d at 466(认定由于律师所在律所已安排关于技术能力与 AI 工具的强制性 CLE 培训,“[i]mposing a sanction of further and additional mandatory education would be redundant”)。
42
参见,例如,Gill v. Everyday Dose, Inc., 2026 WL 3390881, at *5-7 (D. Del. Nov. 9, 2026)(对两名提交 AI 幻觉判例并推卸错误责任的律师处以 3,000 美元罚款);Jimenez-Fogarty v. Fogarty, 831 F. Supp. 3d 253, 266-68 (S.D.N.Y. 2026)(在法院依职权启动的 Rule 11 程序中处以 2,500 美元罚款);Wadsworth v. Walmart Inc., 348 F.R.D. 489, 498-99 (D. Wyo. 2025)(对起草含有伪造判例的文件的律师处以 3,000 美元罚款);Mattox v. Prod. Innovations Rsch., LLC, 807 F. Supp. 3d 1341, 1351-52 (E.D. Okla. 2025)(根据律师各自的责任处以 3,000 美元、2,000 美元和 1,000 美元的累进罚款);Benjamin v. Costco Wholesale Corp., 779 F. Supp. 3d 341, 351 (E.D.N.Y. 2025)(在考虑律师的悔意和补救性 AI 培训后处以 1,000 美元罚款)。
C.A. No. 2025-0426-CDW (LWW)
2026 年 11 月 13 日
第 13 页,共 13 页
要求,他必须在 14 天内亲自向法院支付此款项,
裁决。43 他还必须在完成后提交一份确认付款的宣誓书。
IT IS SO ORDERED.
Sincerely yours,
/s/ Lori W. Will
Lori W. Will
Vice Chancellor
43
参见 Hill, 2026 WL 1146289, at *9(命令受制裁的合伙人“直接且亲自支付此款项(且此款项不得由……律师事务所支付)”);Ct. Ch.
R. 11(c)(4)。
新闻通讯
订阅以接收 Free Law Project 新闻通讯,获取提示和公告。
来源:CourtListener · courtlistener.com