By John W. Amberg

Amberg is a senior counsel with Bryan Cave Leighton Paisner LLP and a former chair and current member of the Los Angeles County Bar Association’s Professional Responsibility and Ethics Committee.  He will be speaking on generative AI at the ABA’s National Legal Malpractice Conference in September.  The views expressed here are his own. Ethics articles are provided regularly by LACBA’s longstanding Professional Responsibility and Ethics Committee.

New technology is seductive, and perhaps none more so than artificial intelligence.  Each year more and more lawyers succumb to the apparent efficiency and economy of generative AI which, with a minimum of prompts, can produce a legal brief replete with case citations and pithy quotes.  And almost weekly, it seems, lawyers are shocked and embarrassed when their cited cases and quotations turn out to be hallucinations - non-existent authorities conjured by software algorithms that are not governed by the ethical duties of competence and candor to the court. California Rule of Professional Conduct 1.1, comment [1] states that a lawyer’s duty of competence includes the duty to keep abreast of changes in the law and its practice, “including the benefits and risks associated with relevant technology.” (Emphasis added.)  Submitting fictitious cases and quotes generated by AI violates Rule 3.1 which requires probable cause for claims and defenses, and Rule 3.3 which requires candor to the court. It is also a potential violation of Rule 8.4 in which the definition of professional misconduct includes “(c) conduct involving dishonesty, fraud, deceit, or reckless or intentional misrepresentation” and “(d) conduct prejudicial to the administration of justice.”  Yet lawyers still do not seem to be getting the message that AI is dangerously fallible.  Consequently, they are paying a price for violating their ethical duties in the form of court-ordered sanctions, referrals for discipline, lost motions, and occasionally, lost jobs.

Recent cases from 2025 provide fresh lessons on the perils of relying on generative AI.  Kohls v. Ellison, __ F. Supp.3d __, 2025 WL 66514 (D. Minn. 2025), concerned the regulation of AI-created “deepfakes.”  Deepfakes are videos, audio recordings or images that are manipulated or altered using AI.  Like many states, Minnesota adopted a law prohibiting the dissemination of deepfakes with the intent to injure a political candidate or influence the result of an election.  Minn. Stat. §609.771. The plaintiffs sued Minnesota Attorney General Keith Ellison to enjoin enforcement of the new law on First Amendment grounds. In opposition to the plaintiffs’ motion for an injunction, Ellison submitted the declarations of two experts to offer background about AI and the dangers of deepfakes to free speech and democracy. One of the State’s experts was Jeff Hancock, professor of communications at Stanford University and director of Stanford’s Social Media Lab.  After the plaintiffs charged that Hancock’s declaration contained fabricated material, Ellison confirmed that the professor had used a generative AI tool – GPT-4o – to draft his declaration, which included citations to two non-existent academic articles and incorrectly cited the authors of a third article. Hancock had failed to verify these false citations. The attorney general claimed his office had no idea that his expert’s declaration contained fake, AI-generated citations and requested leave to file an amended declaration. The District Court denied the request and excluded the Hancock declaration.  In its Order, filed January 10, 2025, the court wrote: “The irony.  Professor Hancock, a credentialed expert on the dangers of AI and misinformation, has fallen victim to the siren call of relying too heavily on AI – in a case that revolves around the dangers of AI, no less.”  Id. at 8.  The court did not fault Hancock for using AI for research. “But when attorneys and experts abdicate their independent judgment and critical thinking skills in favor of ready-made, AI-generated answers, the quality of our legal profession and the Court’s decisional process suffer.” Id. at 9.

The judge reminded the lawyers that Federal Rule of Civil Procedure 11 imposes a personal, non-delegable responsibility on the attorneys to validate the truth and legal reasonableness of the papers filed in the action.  As for the expert, Professor Hancock’s citation to fake, AI-generated sources “shatters his credibility with this Court.”  Id. at 10.

In U.S. v. Hayes, 763 F. Supp.3d 1054, 2025 WL 235531 (E.D. Ca. 2025), an attorney submitted a brief containing AI hallucinations, then proceeded to make matters worse by denying it.  Assistant Federal Defender Andrew Francisco filed a motion to unseal which twice cited and quoted from a non-existent case.  When the government’s opposition brief said the case could not be found, the lawyer claimed the error was inadvertent and that the quoted passage came from another case. This also proved to be untrue. At the hearing, in answer to the Magistrate Judge’s repeated questions, Francisco refused to concede the cited case was not real. The court found that Francisco had submitted a fictitious case which had all the markings of a hallucinated case created by generative AI tools.  When the lawyer also denied using AI to draft the motion, the court found his response was not credible.  In a 28-page order dated January 17, 2025, the Magistrate Judge found Francisco’s persistent misstatements to the court were not inadvertent but knowing and made in bad faith. “Despite being provided multiple opportunities to candidly acknowledge and correct his errors as required under his duty of candor to the court, Mr. Francisco unfortunately failed to do so.”  Id. at 1064.  It ordered him personally to pay sanctions of $1,500 and further ordered the clerk of the court to serve a copy of the order on the bars of California and the District of Columbia where he was admitted, and on all of the other district judges and magistrate judges in the district.  Id. at 1073.

In May 2025, retired Magistrate Judge Michael R. Wilner, appointed as Special Master in Lacey v. State Farm General Insurance Co., __ F. Supp.3d __, 2025 WL 1363069 (C.D. Ca. 2025), ordered the plaintiff’s lawyers at Ellis George and K & L Gates to pay the defendant $31,000 as sanctions for filling a brief containing “bogus AI-generated research.”  Former Los Angeles County District Attorney Jackie Lacey sued her homeowners’ insurer State Farm to recover legal fees incurred in defending her and her husband after he pulled a gun on demonstrators at their house.  The insurer asserted privilege objections to discovery and the parties briefed the propriety of in camera review of the documents. The brief submitted by the plaintiff’s law firms was prepared using AI tools.  The Special Master’s sanctions order concluded that the brief contained numerous false, inaccurate and misleading citations and quotations.  Id. at 2.  Nine of 27 citations were incorrect – two cited authorities did not exist and several quotations were phony. Id. at 3 ¶ 5. No attorney at either law firm had cite-checked or reviewed the research in the brief.  After the Special Master informed the lawyers that he was unable to confirm the accuracy of two authorities, K & L Gates refiled the brief without the two incorrect citations but left the remaining errors in the text.

The Special Master described the brief as a “collective debacle” and faulted both law firms for failing to disclose the use of AI tools, failing to check the validity of the research, and refiling the brief without fixing all of the problems after they were put on notice of errors.  Id. at 7.  “[T]hey had the information and the chance to fix this problem, but didn’t take it.”  He concluded that the lawyers had acted in a manner “tantamount to bad faith” and taken together, their actions demonstrated “reckless conduct with the improper purpose of trying to influence” his analysis, justifying sanctions. Id. at 7-8.

His order manifested the Special Master’s frustration: “Directly put, Plaintiff’s use of AI affirmatively misled me. I read their brief, was persuaded (or at least intrigued) by the authorities . . . only to find that they didn’t exist.  That’s scary.  It almost led to the scarier outcome . . . of including those bogus materials in a judicial order. Strong deterrence is needed to make sure that attorneys don’t succumb to this easy shortcut.”  Id. at 10.

The Special Master’s fear of a judgment based on AI hallucinations came to pass sooner than expected in Shahid v. Esaam, 2025 Ga. App. LEXIS 299 (Ga. Ct. of App. June 30, 2025), a contemporaneous divorce case.  In Shahid, the husband’s lawyer submitted a final judgment containing fictitious citations, which was entered by the court.  Neither the opposing counsel nor the trial court discovered the false citations.  It was not until the court of appeals found new bogus cases in the appellee’s brief that it took a closer look and discovered the fictitious citations in the judgment entered by the lower court.  On appeal, undeterred by the fact that the judgment had relied on two non-existent cases, the same lawyer cited 11 additional cases which the appellate court said were “either hallucinated or have nothing to do with the propositions for which they are cited.“  Id. at __. Acidly, the appellate court noted that the appellee added insult to injury by requesting attorney’s fees on appeal and supporting the request with one of the hallucinated cases. Id.  It vacated the judgment, remanded the case, and ordered the lawyer personally to pay a penalty of $2,500 to the opposing party. Id.

Of course, AI doesn’t always get it wrong.  An AI search tool on President Trump’s social media network Truth Social recently responded that violent crime in Washington, D.C. was substantially down, the Jan. 6, 2021 insurrection was linked to Trump’s “baseless claims of election fraud,” and Barack Obama had the highest favorability rating of living presidents. Drew Harwell, “New chatbox on Trump’s Truth Social platform keeps contradicting him,” The Washington Post, Aug. 10, 2025.  Nevertheless, so long as AI continues to be a beguiling tool, ethical pitfalls will snare lawyers who do not understand its risks. A partner in a Chicago law firm was recently terminated by her firm after she filed a post-trial brief for the Chicago Housing Authority which had been generated by ChatGPT and cited a nonexistent Illinois Supreme Court case. Even though her firm has an express policy against using AI, the former partner told the court that she did not think ChatGPT could create fictitious legal citations. None of the other lawyers at her firm, or her client’s in-house counsel - all of whom also reviewed the brief - caught the error.  Lizzie Kane, “Lawyers for Chicago Housing Authority used ChatGPT to cite nonexistent court case,” Chicago Tribune, July 18, 2025.

The lesson is plain: AI is not a substitute for a lawyer’s critical thinking skills and judgment. To avoid damage to your client, professional embarrassment, and sanctions, do not file anything in a court that a lawyer has not personally researched and verified. If the court questions you about your AI usage, answer truthfully.  And if you have presented fake citations, do not expect to be forgiven.