How To Avoid an AI Horror Story: Verify, Verify, Verify
8.18.2026
Artificial intelligence, like many other forms of technological advancements over the years, has been rapidly adopted by attorneys and courts all over the world, and it is here to stay. By now, we have likely all read about horror stories – hallucinated case citations, incorrectly summarized case decisions, misinterpreted case holdings, cybersecurity risks and ethical violations.[1] This author has penned articles warning about those very risks and proposing some solutions.[2] However, at the end of the day there are many ways in which attorneys and judges can ethically and safely utilize AI and other forms of technology to advance the legal profession and its service to society. It requires application of safe and sound practices that responsible attorneys must oversee and certify. The purpose of this article is to highlight several ways in which the legal profession could address some digital world dangers in everyday practices.
Loss of Privilege or Fifth Amendment Protections Due to Cyber-Hack or Warrant Execution?
Horror Story 1
The reader is no doubt familiar with loss of privilege or confidentiality due to affirmative actions taken by parties interfacing with technology.[3] But what if a search warrant is issued in a criminal matter that addresses electronic materials that fall under the Stored Communications Act (such as submissions and prompts to AI or a chatbot)? Or, what if a malicious actor hacks or unlawfully accesses an account containing privileged or confidential material? Will the protections otherwise afforded by law be lost in those situations?
To begin, if a search warrant is issued, the subject of the investigation – the customer/consumer utilizing the electronic platform – cannot seek to anticipatorily challenge or move to quash the warrant according to United States v. Kim.[4] The service provider is the entity that has limited standing ex ante. The customer or consumer has standing to move to suppress evidence ex post and may seek to do so in protection of their rights at a suppression hearing after return of the warrant.[5] The Kim case, like United States v. Heppner, was from the Southern District of New York, and while such trial court rulings are not binding on any other court or judge and are not precedential, they may be highly persuasive.
Thus, if a chatbot is utilized by a party or client and a search warrant is issued, as in Kim, the material appears thus far to be subject to the initial warrant return (unless quashed on motion by the service provider), but thereafter could be challenged in a suppression hearing by the customer/consumer. In such a case, though, counsel or the party pro se will have to apply or distinguish Heppner, Warner, Morgan, Assini, et al. in an attempt to procure protection for alleged privileged or work-product material and convince the respective court of same.[6]
A related question in the intersecting fields of privilege, constitutional rights and cybersecurity is, therefore, whether privilege or constitutional protections will ultimately be lost if the subject is the victim of a hack, cyberattack or warrant execution or some other action that results in unauthorized viewers accessing protected material? So far, the courts have answered this question in the negative.
In a 2025 decision of the New York State Supreme Court, Appellate Division, Fourth Department, the court ruled that a warrant cannot compel a suspect or target to unlock a mobile device with a fingerprint, as such is equivalent to a testimonial act.[7] While the warrant did not violate the Fourth Amendment, it did violate the Fifth Amendment by requiring the target to unlock the phone – the person thereby acknowledging how to access the phone and demonstrating control over the device and contents. Thereafter, the evidence was suppressed and constitutional protection was not lost.[8]
Given concerns over the prevalence of hacking, spyware, malware and other impermissible electronic intrusions in today’s e-world, one might similarly question whether that access – albeit unauthorized and illegal – compromises the attorney-client privilege. The answer to that, at least in New York, also still appears to be “no,” according to the court in C.C. v A.R.[9] Pursuant to the court’s holding in C.C., if the cyber-attacker is another party to the case at issue, they could face very strict sanctions:
“This appeal examines the proper sanction in a matrimonial litigation where one party installed spyware on the other party’s phone, invoked Fifth Amendment protections on the issue, and intentionally destroyed evidence as to what the spyware intercepted. The [trial court] properly determined, under the circumstances here, that it is proper to infer that the plaintiff violated the defendant’s attorney-client privilege, and that the appropriate sanction is to strike the causes of action in the complaint relating to the financial issues of the case other than child support. … Here, where the plaintiff violated the defendant’s attorney-client privilege, the preclusion of documents is not an appropriate alternative sanction because the plaintiff will always be privy to any litigation strategy of the defendant that he gained through secretly intercepting her privileged communications. … As the plaintiff “chose the course of action that exacerbated the harm”…, the defendant has been seriously compromised in defending herself in this divorce action.”[10]
Therefore, as a practicing attorney, be cautious, use proper security protocols, remain vigilant and take continuing legal education courses to update skills and knowledge in cybersecurity, and advise clients to take some of the same safeguarding and preventive measures. Attorneys may wish to expand their representation letters or retainer agreements, advising their clients/entities in writing regarding the above case law and court holdings; counseling them about whether it would be best to avoid chatbot/AI use; and advising regarding practical limits. Having the client countersign the letter, agreement or writing, and maintaining a copy in the file will document that counsel has advised the client or corporate entity (in the case of in-house counsel), warned them of the risks, and acted to mitigate potential future loss of privilege from what would otherwise have been the client’s unguided utilization of technology to research legal issues prior to, or in place of, discussing the same with their attorney.
Then take some comfort in recognizing that unauthorized and illegal access to privileged communications occurring from spyware, malware or warrant executions that violate the Fifth Amendment thus far do not appear to compromise privilege or the right against self-incrimination in the eyes of the courts. However, beware: As the world and the courts become ever more familiar with the risks and benefits of technology, and as attorneys are expected to be more familiar with those and responsive security protocols are implemented, courts might grow to be less protective and tolerant of breaches that cause loss of privileged or confidential information. It is vital, then, that an attorney’s protocols are robust, counseling of clients is thorough, and overall, the attorney demonstrates competence and sound practices.[11]
Ensure You Have Properly Programmed Any AI System That Assists With Hiring
Horror Story 2
AI is utilized as a hiring manager assistant to review numerous job applications in order to save time and money and other personnel resources at the hiring entity. AI then makes “decisions” utilizing biased programming. Could this result in liability and/or ethics sanctions for discriminatory decision-making in violation of law?
As AI technology has expanded, so have uses of AI in most areas of our society. This includes utilizing AI resources to assist in culling through sometimes myriad submissions in response to job postings – in essence, making AI the first-level human resources manager regarding résumé reviews. However, beware of two concomitant concerns – “automation bias” and potentially discriminatory hiring practices.
Automation bias is where humans presume that technology and AI are, basically, infallible, and we therefore accept and confirm whatever they tell us or show us largely without question[12] (thus also helping to explain why we are seeing so many cases involving hallucinated citations in briefing). This was shown to be a major problem in the law when an individual was falsely accused of being a thief by AI facial recognition when he was 30 miles away (confirmed by alibi evidence).[13] The alibi evidence was almost pushed aside because the AI facial recognition was taken as gospel. However, the defendant, his attorneys and experts questioned the AI, and he was exonerated.[14]
Additionally, we are all familiar with the saying “garbage in, garbage out” – referring here to how well the AI platform is programmed; that is, the level of quality in output one can expect to receive. Thus, be careful if your hiring manager is also an AI platform – even if only for the initial round of résumé reviews. Legal commentators, attorneys and other experts have been discussing the risks associated with AI programs used to screen applicants in a hiring process. Because AI first utilizes the data with which it is programmed and then learns, there is concern that old methods and data, and bias in the system, could result in bias during screening and hiring phases.[15]
Some jurisdictions, such as New York City, have instituted laws to require system audits and to punish those employers or employment agencies whose AI systems are found to include illegal bias.[16]
Attorneys in New York should also be familiar with the provision in the Rules of Professional Conduct that states: “A lawyer or law firm shall not: … (g) unlawfully discriminate in the practice of law, including in hiring, promoting or otherwise determining conditions of employment on the basis of age, race, creed, color, national origin, sex, disability, marital status or sexual orientation. …”[17] While there has been some debate concerning the extent of 8.4(g) to an attorney’s arguments in representation of a client,[18] attorneys still must exercise caution when the potential for discrimination in hiring within their firm or agency exists.[19]
If law firms, public agencies or client companies use AI to assist in hiring, they should be aware of the potential risks related to programming and bias – and the potential legal or ethical violations. Here again, attorneys should be familiar with all risks and benefits of technology under the New York Rules of Professional Conduct 1.1 and should ensure that their own practices within their firms or agencies comply with the requirements of law and ethics, while also counseling their clients to ensure the same under the relevant laws and regulations. Audits and random checks of AI platforms (if utilized) – their programing and outputs – should be regularly and routinely conducted to ensure necessary compliance and best practices.
Cite-Check All Briefing, Whether From an Associate, Co-Counsel or Opposing Party
Horror Story 3
An attorney is working with co-counsel in his or her own firm or agency, in addition to local counsel or pro hac vice counsel. The attorney is one of the counselors of record in litigation. Receiving draft briefing from co-counsel, and believing he or she is capable, the attorney in question does not cite-check, and directs co-counsel to sign it and file it with the court. The briefing turns out to contain AI hallucinations, and violates N.Y. Rules for Professional Conduct 3.3, among other rules. Is the otherwise non-offending attorney liable for violations?
Attorneys who do everything right in their own practice and in their own submissions may believe that they have insulated themselves from reproach and protected themselves from ethical criticism. However, that is not necessarily the end of the story. Recently, even some “non-offending” attorneys are being criticized by courts for not being alert to errors in briefing, although contained in submissions not made by that attorney – such as when an attorney fails to cite-check a response or reply brief (depending on the motion and court) from an opposing party on a motion. Normally, absent leave of court, a response to a reply brief (or sur-reply) would not be filed, and the motion would be deemed fully submitted.[20] It might be tempting for an attorney to accept that final reply briefing, and await the court’s ruling without doing more (and thus without charging the client for additional review and cite-checking of a brief to which no response would be filed, or without taking time away from another taxpayer-funded matter in the case of an agency). However, courts have begun to criticize non-offending attorneys who wait for judges to be the “last line of defense” – although not yet sanctioning them outright, since technically they did not submit in violation of the rules.[21]
If attorneys are serving as co-counsel, pro hac vice counsel, or local counsel, it will behoove them to cite-check and verify all statements and case citations in any briefing prepared by colleagues, or to which they sign on, or is submitted over their signature or in a case in which they are one of the counselors of record, for risk of being swept up in future sanctions for rules violations.
For example, in Billups v. Louisville Municipal School District,[22] a Mississippi federal district court sanctioned an attorney serving as co-counsel within a firm for not checking and catching incorrect briefing containing AI hallucinations and misstatements by their co-counsel for whom the court found they were a supervisory attorney; the supervisory counsel’s name had appeared on the initial complaint.[23] Furthermore, that same court then issued a challenge to the opposing party and its counsel – who to that point had not been a direct offending party:
“The Court also observes that the Defendant, … could have flagged the fictious citation and misrepresentation of case law in a reply brief or supplemental filing. … The Court takes this opportunity to issue a charge. Going forward, the Court expects all parties to assist in maintaining the integrity of the judicial process and to be diligent in flagging AI misuse. ‘[O]therwise, the risk is too great that such errors will persist undetected, potentially leading to an outcome unsupported by law.’”[24]
Separately, another federal district court in Mississippi addressed an unusual situation where counsel for both parties submitted a briefing that had utilized AI resources, but they did not verify their submissions, and it contained non-existent or incorrect citations.[25] The Withers court noted that, by submitting fake, unverified cases, counsel had obviously not read (or had not attempted to read, in any event) those non-existent cases. Such are Rule 11 violations in federal court (as well as ethics violations).[26] The court disqualified and barred both sets of counsel and ordered one of them “to attend a CLE on artificial intelligence with an ethics component addressing the obligations of counsel regarding the use of artificial intelligence in the legal field and submit proof of attendance within 60 days from the date of this Order.”[27] Monetary fines were issued, as well, and the court directed that a copy of the order be sent to the respective state bars of the offending attorneys.[28]
Finally, there is a particularly egregious case, In re Domestic Partnership of Torres Campos & Munoz,[29] out of California. In Munoz, one of the party’s counsel cited fictitious case law in a submission. Compounding the problem, the opposing party’s counsel thereafter relied on the fictitious case law, and re-cited it in their own submission to court. The trial court then relied on that same fictitious case law and issued an order citing same. The appellate court admonished all involved in the case, held that the trial court abused its discretion by citing fictitious case law as support for the order, sanctioned counsel, and ruled that the appellant had thereby forfeited their claim.[30]
Hence, our last takeaway – attorneys should be certain to cite-check any submissions to courts or opposing parties that bear their signature, on which their names appear, or in any litigation in which they have taken part in any way – including submissions prepared by attorneys on their own side of the case, as well as on the opposing side.[31] Said attorneys have supervisory and ethical responsibilities to do so.[32] Every single citation, case summary, digest and argument should be sourced and confirmed, in one’s own briefing and all opponents’ briefing – including all reply briefs to which one might not otherwise respond. If nothing else, failure to do so by an otherwise non-offending party/counsel could at least result in a reprimand from the court, a challenge from the court or possibly worse – such as loss of a sanctions award that could have otherwise been received by that non-offending party.[33]
It is both a startling and exciting world out there, as technology continues its rapid advances – not just in the law, but in all parts of society. Yet, if attorneys take care to exercise oversight and adopt a few cautionary and ethical practices, harnessing technology for use in the legal profession might be quite attainable and may also be quite advantageous.
Michael L. Fox is the acting dean and chairperson of the School of Business and associate professor of business law at Mount Saint Mary College. He is also an adjunct professor at Western Connecticut State University. He is a former NYSBA vice president (Ninth Judicial District), section chair and committee chair and former member of the NYSBA Executive Committee and House of Delegates.
Endnotes:
[1] Cassata v. Michael Macrina Architect, P.C., 250 N.Y.S.3d 778 (Sup. Ct. Suffolk Cty. 2026); Deutsche Bank Nat’l Tr. Co. v. LeTennier, 250 N.Y.S.3d 260 (3d Dep’t 2026); Grymes Dev. Co. v Fodera, 88 Misc. 3d 767, 245 N.Y.S.3d 863 (Sup. Ct. Richmond Cty. 2025); Gully v. Varghese, 89 Misc. 3d 1208(A), 2026 WL 1657540 (table) (Sup. Ct. Albany Cty. 2026) (citing, inter alia, Grymes); Landberg v. City of New York, ___N.Y.S.3d___, 2026 WL 1792016 (2d Dep’t June 23, 2026) (discussing, inter alia, LeTennier and Julien v. Arthur, ___N.Y.S.3d___, 2026 WL 1476261 (2d Dep’t May 27, 2026), and issuing hefty sanction against attorney for AI hallucinations and violations of RPC).
[2] Michael L. Fox, AI & Hallucinations Cause Ongoing Concerns; The Solution May Be Amended Ethics Rules, N.Y. L.J. at 4, 7 (June 5, 2026); Cynthia Feathers & Michael L. Fox, Core Ethics for Appellate Lawyers in the Age of Gen AI, N.Y. L.J. at pp. 4 & 8 (Dec. 23, 2025).
[3] See, e.g., United States v. Heppner, 820 F.Supp.3d 292 (S.D.N.Y. 2026); but cf. Morgan v. V2X, Inc., 2026 WL 864223 (D. Colo. Mar. 30, 2026) (distinguishing Heppner because it was a criminal matter versus the civil case before the Court, and in Heppner the party “acted entirely apart from his lawyer”). See also Assini v. Hayward, ___ N.Y.S.3d ___, 2026 WL 1677232 (Sup. Ct. Nassau Cty. June 4, 2026) (citing Morgan and Heppner, also distinguishing rule in criminal cases versus civil cases; defendant warned about complying with new 22 N.Y.C.R.R. Part 161, so as to avoid “use of AI frustrat[ing] the litigation,” and going unfettered; citing Augustin v. Formula 3 Brook. Inc., 86 Misc. 3d 1236(A), 236 N.Y.S.3d 586 (Sup. Ct. Kings Cty. 2025)).
[4] United States v. Kim, 2026 WL 1785995 (S.D.N.Y. June 22, 2026).
[5] Kim, 2026 WL 1785995, at *1 (citing, inter alia, United States v. Grubbs, 547 U.S. 90, 99 (2006); Warshak v. United States, 532 F.3d 521, 523 (6th Cir. 2008); but not citing Heppner).
[6] See supra note 3.
[7] See Brian Lee, New York Appellate Court Rules Search Warrant Cannot Compel Cellphone Unlocking by Fingerprint, N.Y.L.J., July 2, 2025, at 1; People v. Manganiello, 242 A.D.3d 41 (4th Dep’t 2025).
[8] Id.; see also United States v. Brown, 125 F.4th 1186 (D.C. Cir. 2024) (on the specific issue of mobile device and Fifth Amendment).
[9] C.C. v. A.R., 192 A.D.3d 654, 143 N.Y.S.3d 404 (2d Dep’t 2021).
[10] Id. at 654-55.
[11] See, inter alia, N.Y. R.P.C. 1.1 & Comment 8, 1.2, 1.3, 1.4, & 5.3.
[12] See generally, Lauren Kahn et al., AI Safety and Automation Bias, The Downside of Human-in-the-Loop, Center for Security & Emerging Technology (Nov. 2024), https://cset.georgetown.edu/publication/ai-safety-and-automation-bias/.
[13] See Joel Cohen & Douglas Nadjari, Fighting the Demons of AI: New Battlegrounds, N.Y.L.J. Jan. 5, 2026, at 3 (discussing Complaint & Jury Demand, Parks v. McCormac, No. PAS-L-003672-20 (N.J. Super. Ct. Law Div. Passaic Cty.)).
[14] Id.
[15] See Erika Collins & Ari Rutledge, AI Programs Used by Employers May Lead to Title VII Discrimination Claims, N.Y.L.J., Aug. 11, 2023, at 7.
[16] See N.Y.C. Admin. Code §§ 20-870 – 20-874 (2023); see also Isha Marathe, As AI’s Role in Hiring Grows, Focus Turns to Understanding Its Decision-Making, N.Y.L.J., Oct. 21, 2022, at 5; Brad Raboin, et al., NYC’s AEDT: A Growing Regulatory Arena and Efforts Seeking Clarity, N.Y.L.J., Dec. 8, 2022, at 4.
[17] N.Y. Rules of Prof’l Conduct 8.4(g).
[18] See, e.g., inter alia, Bruce A. Green & Rebecca Roiphe, ABA Model Rule 8.4(g), Discriminatory Speech, and the First Amendment, 50 Hofstra L. Rev. 543 (2022).
[19] See, e.g., N.Y. City Bar Comm. On Prof’l Ethics, Formal Op. 2020-4 (2020).
[20] See Local Rules of the United States District Courts for the Southern and Eastern Districts of New York, Local Civil Rule 6.1 (2026); see also Part 202. Uniform Civil Rules for the Supreme Court and the County Court, Section 202.8-c Sur-Reply and Post-Submission Papers (2026); see also U.S. District Court for the District of Maryland, Responses and Replies, https://www.mdd.uscourts.gov/content/responses-and-replies (“A surreply is not permitted to be filed without leave of court. See Local Rules 105.2.a”) (last visited July 8, 2026). Prohibitions on sur-replies can also be found in the Individual Practices of certain judges – see Individual Rules of Practice of Judge Cathy Seibel [S.D.N.Y.] Rule 2.B.i; Individual Practices in Civil Cases, Jennifer E. Willis, United States Magistrate Judge [S.D.N.Y.], Rule II.A.
[21] See Nuvola LLC v. Wright, 2025 WL 4083694 (Minn. Dist. Ct. Hennepin Cty. Nov. 21, 2025) (court did not sanction counsel, because they were not the ones to cite nonexistent cases, but was bothered by fact that counsel left court to be “last line of defense”); Dec v. Mullin, 171 F.4th 940, 947 (7th Cir. 2026) (“Though Dec’s counsel described her efforts to mitigate future errors, she should have been more careful. Still, she took responsibility and showed contrition. That opposing counsel also failed to catch these errors and bring them to our attention also gives us pause, albeit to a lesser degree”) (emphasis added).
[22] Billups v. Louisville Municipal School District, Civ. Action No. 1:24-CV-74-SA-RP (N.D. Miss. Dec. 19, 2025).
[23] Id.
[24] Id. at *8 (citing Ferris v. Amazon.com Servs., LLC, 778 F. Supp. 3d 879, 880-81 (N.D. Miss. 2025); Elizondo v. City of Laredo, 2025 WL 2071072, at *3 (S.D. Tex. July 23, 2025)).
[25] See Withers v. City of Aberdeen, 2026 WL 1653484 (N.D. Miss. June 8, 2026).
[26] Id.
[27] Id.
[28] Id.
[29] Dom. P’ship of Torres Campos & Munoz, 118 Cal.App.5th 1112, 342 Cal.Rptr.3d 227 (Cal. Ct. App. 4th Dist. 2026).
[30] Id.
[31] Note – this does not mean that an attorney must go on a research database and conduct their own first-hand research all over again, possibly at a senior attorney or partner-level billing rate, and after the client (in private firms or corporations) or the taxpayers (for public agencies) have already been billed for the research and time drafting and editing the submission. The attorney needing to cite-check as a supervisory matter could – if not already utilized as a potential course of action – request that the draft submission be given to them with copies of all cited cases, statutes and regulations pulled from verifiable and acknowledged research sources or databases – such that the attorney upon review is certain the cited material exists, is real and is properly characterized in the submission to the court. The same may be employed by judges in receiving draft opinions and orders from law clerks, if not already the practice in chambers.
[32] 22 N.Y.C.R.R. 130-1.1; N.Y. Rules of Prof’l Conduct 3.3, 4.1, 5.3.
[33] Noland v. Land of the Free, L.P., 114 Cal. App. 5th 426, 336 Cal. Rptr. 3d 897 (Cal. Ct. App. 2d Dist. 2025) (sanctions ordered payable to court, but no sanctions payable to opposing counsel because they failed to alert court to fabricated citations in the other counsel’s brief).
