Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence within the justice system. It supports my work in teaching, lecturing, and writing about AI and the law and is published to promote my practice. Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI Law.
Introduction
The first week of September 2025 is not looking good. There are already eight reported incidents of AI Hallucinations and/or fabricated citations. Last weeks news letter can be found here.
Just a reminder for both regular and new readers: I begin with a basic search using recognised, publicly available legal databases and blogs. I then use AI agents to identify any further cases I may have missed. After that, I review the links and write about anything I believe readers of this blog may find useful. This process takes time, but I enjoy the subject. That said, mistakes can slip through, so please check the sources directly. This blog offers commentary, not legal advice.
Davis v. Marion County Superior Court Juvenile Detention Center (USA)
No. 1:24-cv-01918-JRS-MJD (S.D. Ind., 2 September 2025)
On 31 July 2025, an attorney filed a response brief to Defendants’ Motion to Compel. In the brief, the attorney included two citations that he conceded did not exist. Firstly, the brief contained a quotation and citation that was:
” a rather bizarre amalgamation of several cases. [] is a cause number that exists; however, it does not involve either a party named Perry or the City of Indianapolis, and it was never assigned to either Judge Sweeney (JRS) or Magistrate Judge Pryor (DLP). There have been cases with Perry and the City of Indianapolis as parties, but none of them have any opinions on Westlaw that relate to discovery. 2019 WL 2088435 is a real citation, but it is for the case … from the New York Superior Court. It is, at least,
a decision dated May 13, 2019, but it has nothing to do with the burden in a discovery dispute…”
Secondly, the brief contained a further representation, citation, and quote, but:
“The cited case does not exist. There is a cate with citation…, but that does not involve tax returns”.
An order to show cause was issued directing the attorney to appear and show cause why he should not be sanctioned. The attorney accepted responsibility for the errors, but:
“… he took the position that the main reason for the errors in his brief was the short deadline (three days) he was given to file it. He explained that, due to the short timeframe and his busy schedule, he asked his paralegal (who once was, but is not currently, a licensed attorney) to draft the brief, and [attorney] did not have time to carefully review the paralegal’s draft before filing it…”
The magistrate judge found that the attorney failed to take even the most basic of actions, and therefore did not catch that the brief contained citations that did not exist. The judge observed:
“The most logical explanation for the citation to non-existent authority is, of course, the use of generative AI to conduct legal research and/or draft the brief. The issue of “hallucinated case[s] created by generative artificial intelligence (AI) tools such as ChatGPT and Google Bard” has been “widely discussed by courts
grappling with fictitious legal citations and reported by national news outlets.”
That was denied by the attorney, who represented at the hearing that the paralegal did not use generative AI to aid in the drafting of the brief, but had used a research product. The judge explained:
“…whether generative AI was or was not used to draft the brief is not particularly relevant to the analysis. There is nothing fundamentally improper in the use of AI tools to draft a brief. Rather, it is counsel’s abdication of his responsibility to ensure that the information he provided to the Court was accurate that is the basis for the sanctions recommended.”
The Judge set out the legal framework adding:
“the advent of modern legal research tools implementing features…has enabled attorneys to easily fulfill this basic duty, and there is simply no reason for an attorney to fail to do so. Such has been the view for decades: “It is really inexcusable for any lawyer to fail, as a matter of routine, to Shepardize all cited cases (a process that has been made much simpler today than it was in the past, given the facility for doing so… Confirming that a case is good law is a basic, routine matter and something that is expected from a practicing attorney. As noted in the case of an expert witness, an individual’s “citation to fake, AI-generated sources . . . shatters his credibility.”…The same is true even if the fake citations were generated without the knowing use of AI”
The magistrate judge considered that the court need not make any finding as to whether the lawyer actually used generative AI to draft any portion of his motion and reply, including the fictitious case and quotation. Citing non-existent case law or misrepresenting the holdings of a case is making a false statement to a court. It does not matter if generative AI told you so. Further, the fact the attorney did not intend to deceive the court did not excuse the failure. The magistrate judge found this violated Rule 11 and recommended a personal 7,500 dollar sanction, together with a referral for discipline.
Lockwood v. ICBC (Canada)
2025 BCCRT 1227 (BC Civil Resolution Tribunal, 3 September 2025)
In this accident benefits dispute, a self-represented claimant said:
“… she reserved the right to submit evidence obtained through this request as late evidence in this CRT dispute, or to request a reconsideration under “CRT Rule 15” should my decision be issued before the records were received….Through staff, I advised the parties that CRT rule 15 does not exist, and there was no procedure for the CRT to reconsider its own decisions. I also said that in the absence of a request from [applicant] to pause the proceedings, I would proceed with adjudicating the claims before me….”
Later, the tribunal observed:
“…This would make it a new application rather than an appeal or request for reconsideration. [applicant] responded by citing non-existent or inapplicable sections of regulations, likely a result of relying on artificial intelligence, as she did in her confusing reply submissions in this dispute…”
The tribunal did not appear to have imposed any sanction for this and moved on to determine the case.
Pete v. Houston Methodist Hospital (USA)
No. 1:25-cv-00273-MJT-CLS (E.D. Tex., 3 September 2025)
A pro se plaintiff sought a default judgment against a hospital, but there were concerns about whether service had been accomplished:
“Plaintiff appeared in her pro se capacity. When questioned about attorney Alexandra Smith, Plaintiff admitted she never met Ms. Smith and that Ms. Smith did not represent her. Plaintiff then admitted that she used Artificial Intelligence to prepare the motion. Judge Stetson advised Plaintiff that the State Bar of Texas has no record of an Alexandra J. Smith from Houston with this bar number and cautioned Plaintiff
about using AI to prepare legal documents as she would be held responsible for the content it
creates.”
The court then looked closer at the issue and gave a warning:
“At the hearing, Plaintiff admitted that she had not served the Defendant and that she was not entitled to a default judgment. Judge Stetson advised Plaintiff that she would issue a report and recommendation denying the motion for default judgment. She also cautioned the Plaintiff that further filings containing false facts or AI-generated false caselaw could result in sanctions.”
The plaintiff then objected to the report and recommendation on two grounds. The first was overruled. The second concerned a 30-day delay from the hearing to the report and was “not well-taken”, but in citing two cases in support of her argument:
“The Court cannot locate either of these cases in either the Westlaw database using the citations provided or in the Eastern District of Texas filing system. Despite the admonitions of Judge Stetson, these cases may be fake. Plaintiff has shown no such prejudice on the part of the Magistrate Judge.”
The court overruled the plaintiff’s objections, adopted the report and recommendation and the plaintiff was:
“…ORDERED TO SHOW CAUSE why she should not be sanctioned under FED. R. CIV. P. 11 for filing motions that contain false statements and/or fake caselaw….”
Steven E. Hobbs, Sr. v. Igor Goncharko, et al. (USA)
No. 25 CV 3398 (N.D. Ill., 3 September 2025)
In this housing case, a self-represented tenant alleged discrimination by his landlords. In ruling on motions to dismiss, the court briefly noted that the tenant twice cited a case:
“…which neither defense counsel nor the court can locate, for the proposition that “[b]rokers can be liable under FHA for discriminatory leasing practices.” The ghost citation aside, he is correct that discriminatory leasing practices are actionable…d But he is not alleging leasing discrimination on Coldwell Banker’s part—only that it facilitated a transaction that, eventually, led to discriminatory treatment by other actors. This is insufficient to survive a motion to dismiss under Rule 12(b)(6). The § 3604(b) claim against Coldwell Banker is therefore dismissed.”
No sanction was imposed, but a footnote warned that:
“Federal Rule of Civil Procedure 11 applies to pro se litigants, and sanctions may result from such conduct, especially if the citation to [the questionable case] was not merely a typographical or citation error but instead referred to a non-existent case. By presenting a pleading, written motion, or other paper to the court, an unrepresented party acknowledges they will be held responsible for its contents”
Mark Khoury v. Nira Kooij (Australia)
A probate dispute between siblings took an unusual turn: the self-represented brother’s submissions conculded with certain contentions which the court considered in detail:
“…He refers to [redacted]. The case which is reported at [2017] QSC 105 is Attorney-General v Guy which is a decision concerning the Dangerous Prisoners (Sexual Offenders) Act 2003. On the Queensland Judgments website there is no case with the name [redacted]. Further, he purports to quote from that decision to this effect: [redacted]. I have been unable to find that series of words used in any reported or unreported decision in Queensland or in cases able to be searched in Lexis Advance.”
…The arguments that disclosure is premature, or that no claim is yet made, must be dismissed. He relies upon Macedonian Orthodox Community Church St Petka v His Eminence Petar [2008] HCA 42; (2008) 237 CLR 66 at [123] for that proposition. That case does exist. But it is not authority for the point advanced by Mr Khoury and paragraph [123] has nothing to do with the point advanced. He goes on to cite two other cases – neither of which exist.
(c) Concealing costs from a co-executor breaches mutual fiduciary duties. He relies on [redacted]. That case does not exist. There is a decision with that citation but it has no relevance. The same applies to his reference to [redacted]. The decision with that citation is a case which concerns the construction and interpretation of contracts. He also refers to s 54(1)(a) of the Succession Act 1981. Section 54(1) does exist. There is no sub paragraph (a) of that section and the section relates to the protection of persons acting informally.
The judge then considered the most likely reason for these false citations and reiterated the practice of the Federal Court:
[16] I will assume for present purposes that [applicant] has not deliberately concocted authorities in an attempt to mislead the court. It is more likely that he has relied upon an AI chatbot and that these contentions and citations are the result of some form of AI hallucination. Whatever the source of these citations, there is no authority for the propositions he advances.
[17] When these reasons are published generally, I will follow the practice used by the Federal Court in Luck v Secretary, Services Australia. The names and citations of the non-existent case names and the reference to non-existent extracts will be redacted so that the false information is not propagated further by artificial intelligence systems having access to these reasons.
The application was dismissed.
Re X Corp (Canada)
2025 BCCRT 1228 (4 September 2025)
In March 2025, a protection order was granted under Intimate Images Protection Act (IIPA) section 5 in file IS-1-2025-000905. The applicant in that unpublished decision (TR) alleged the respondent did not comply with the protection order. This was denied by the respondent who claimed they did comply. However, the court ordered the respondent to pay the maximum $100,000 administrative penalty to the Minister of Finance and then turned to consider whether TR was entitled to compensaation for her time spent on this dispute:
“44. TR asks for compensation for her time spent on this dispute. IIPO rule 11.4(3) says that the CRT will not order a party to pay another party for time spent on a dispute except in extraordinary circumstances. TR says that there have been repeated violations of the protection order on [respondent] as posters continue to upload copies of the intimate image. However, it is not [respondent’s[ fault that people continue to post the intimate image. She also says that [respondent] should have accepted service through its online portal, and its failure to do so wasted her time. However, the IIPO rules dictated how TR needed to serve [respondent], and I find [respondent] had no legal obligation to accept service through its online portal.
45. Also, TR’s reply submissions were drafted with the assistance of artificial intelligence, a fact TR disclosed. It is becoming increasingly common for parties to use artificial intelligence, often resulting in inaccurate or misleading submissions. In a previous unpublished decision, I warned TR about relying artificial intelligence for these reasons. Unfortunately, her reply submissions suffer from the same flaws. For example, the first case cited in her reply submissions is real. The submissions then reproduce a block quote from paragraph 150 of that case. However, the case is only 124 paragraphs long and the quote does not exist. Her submissions continue to confidently cite the case for legal propositions it does not discuss. So, whether knowingly or negligently, TR provided false and misleading submissions. Given that compensation for time spent is an extraordinary and discretionary remedy, I find this is enough reason alone to dismiss her compensation claim.“
Turner v. Garrels (USA)
No. 24-0895 (Iowa Ct. App., 4 September 2025)
A father appealed the district court’s custody order granting the mother physical care and sole legal custody to their child. In addition to arguments regarding the custody and physical care issues, he argued that Iowa Code section 610.1 (2022), which provides the fee for ordering the transcript of the district court proceedings cannot be deferred, is unconstitutional and that requiring him to pay the transcript fee violates his constitutional rights. Because he waived all issues he has raised on appeal, the court affirmed the district court’s order.
However, the court noted in a footnote that the father:
“… quoted at least one nonexistent case in his judicial-notice motion. It thus appears that [father] may have misused an artificial intelligence tool when preparing the motion...”
The court warned:
So we once again “stress that selfrepresented litigants and attorneys alike have a duty to independently verify the authenticity and veracity of all sources and assertations when relying on artificial intelligence tools to prepare trial or appellate court filings.” Luke v. State, No. 24-1421, 2025 WL 2237311, at *1 (Iowa Ct. App. Aug. 6, 2025). Although a pro se litigant is not subject to the same ethical duties as an attorney, we expect attorneys and pro se litigants to “all . . . act with equal competence.” See Kubik v. Burk, 540 N.W.2d 60, 63 (Iowa Ct. App. 1995)…”
N.Z. et al. v. Fenix International Ltd. (OnlyFans) (USA)
No. 8:24-cv-01655-FWS-SSC (C.D. Cal., 4 September 2025)
OnlyFans users (as anonymous “N.Z., et al.”) sued the platform’s owner for failing to protect them from illicit content. This claim requires a detailed post, as it has a long history and raises important issues. For now, the following “prelimanry statement” sets out the allegations of AI hallucinations:
“Plaintiffs’ Opposition (“Opp’n”) cites 18 cases in attempting to argue that the Court should decline to partially reconsider its FNC Order, or certify an interlocutory appeal, in response to the California Supreme Court’s recent decision … As discussed in detail below and in the accompanying Declaration …Plaintiffs attributed false, AI-hallucinated quotations or holdings to at least 11 of those cases. This is the third distinct filing over a monthlong period in which Plaintiffs have used non-existent quotations to attempt to defeat Fenix’s requests for relief. ..This pattern of submitting false, AI-generated law is an “abuse of the judicial system” that harasses Fenix and wastes the Court’s time and resources. Cooter & Gell v. Hartmarx Corp., 496 U.S. 384, 398 (1990); Mata v. Avianca, Inc., 678 F.Supp.3d 443, 461 (S.D.N.Y. 2023) (“An attempt to persuade a court or oppose an adversary by relying on fake opinions is an abuse of the adversary system.”). It must be stopped. The Court should disregard Plaintiffs’ latest tainted efforts and grant the Motion.”
Judge Slaughter has issued an Order to Show Cause in chambers addressing AI‑generated content in the opposition briefs and calendared a sanctions hearing to determine whether penalties or other remedies should issue. The sanctions hearing is set for 25 September 2025
Comment
It is concerning that only one week into September we already have eight cases to consider. This is also the first time I have seen a judge describe a pattern of submitting false, AI-generated law as harassment. I will need to give that further consideration and will be following the hearing on 25 September 2025 with great interest.
Let’s hope we have fewer incidents next week.
I’d be very interested to hear your thoughts on false citations/AI hallucinations wherever you practice especially if you are aware of any reported incidents I may have missed. If you’ve found this article useful please consider subscribing to my Substack newsletter, where I regularly share broader legal commentary. Many of you regularly read these articles, which is great, but comparatively few subscribe, subscribe here.
Final Comment From o5
This is where I offer a premium model the chance to comment or critique the preceding discussion. Here is its response:
“Substantively, your roundup captures a clear cross‑jurisdictional convergence: courts are not condemning the use of AI per se but are sanctioning the abdication of the non‑delegable duty to verify authorities and facts, with Rule 11 and analogous powers applied to both counsel and litigants in person; that point is especially well drawn in the Indiana and Iowa matters, and the Canadian CRT example shows how AI‑tainted submissions can carry collateral consequences even where the main sanction arises elsewhere. The Queensland redaction approach is significant as a remedial policy aimed at preventing the seeding of false material into public datasets and future models, and it would be worth tracking whether other courts adopt similar prophylactics. One nuance to flag is the OnlyFans docket: the “harassment” characterisation appears to be an adversarial allegation that will be tested at the sanctions hearing rather than a judicial finding at this stage, so treating it as a developing issue is prudent. A valuable extension of your analysis would be to categorise incidents by error type, state of mind and sanctioning basis, and to monitor emerging gatekeeping measures such as certification requirements, hyperlinking to sources, or mandatory representations about the use and validation of AI in filings.”
